Your own subscriptionsImproved
Before your first team starts, setup and rig doctor say which logins it needs, and the operator lets you choose if one is missing.
Every change in each OpenRig release, with the pull request behind it.
Latest release
OpenRig 0.7.20 waits out a Claude usage limit and a lost tmux server. When a parked task's wake is refused because a Claude seat has hit its usage limit, OpenRig records why and retries at the reset Claude states; a Codex seat states no reset, so it keeps its ordinary retry. When the tmux server behind a seat goes away, the daemon pauses automatic delivery to that server's seats, checks back without starting the server, and offers the owed work again once it answers; rig status, rig ps and rig doctor report the outage. The rigs skill now opens with the consent rules your coding agent must keep, and tells it to read a handed-over task's result with rig queue show <id> --full. A workflow instance can be bound to one pod, and an opt-in throttle limits how often a Claude seat's status line is collected. Many fixes come from people using OpenRig.
rig seat status shows a seat left at a bare shell as needing attention.When a parked row's wake is refused at a Claude seat that has reached its usage limit, the wake's record says so and retries at the stated reset; Codex states no reset, so a Codex seat's wake keeps its ordinary retry. A Codex seat's generation record is read from its declared home. rig seat status reads a seat left at a bare shell as needing attention, as rig ps does. rig daemon stop acts on a start reservation's child only after confirming the reservation still names it. Queue-only manual compaction prepares under the lifecycle lease. The draft guard recognizes its own collapsed paste. These change how recovery and delivery decide. This release doesn't claim a measured recovery improvement on a running team.
rig status, rig ps and rig doctor report the outage. A community contribution.When the tmux server a seat depends on goes away, the daemon records it once and holds automatic delivery to that server's seats. It probes read-only every 5 s, backing off to 60 s, without starting the server, and offers owed work again when the server answers. An explicit rig send is still attempted. rig status, rig ps and rig doctor report the outage.
rigs skill now opens with the consent rules your coding agent must keep from its first handoff, and tells it to read a handed-over task's result with rig queue show <id> --full.The rigs skill opens with the consent rules an agent must hold from its first handoff, even if it reads nothing else, and points to step 2 for the full rules. It hands work over with a return route: most of the team's questions come back to the person's conversation, and the result is in the handed-over task's resolution, which the agent reads with rig queue show <id> --full. This release ships that guidance; it doesn't claim a measured change in how agents behave. rig setup's next steps and the install reference say the same.
rig daemon stop prints Daemon stopped.Run the upgrade in the environment that owns your daemon, keeping the same OPENRIG_HOME (legacy RIGGED_HOME) if you set it. Stop the daemon, back up its database, install the patch, then restart it. Your agents keep running; do not use rig down for this upgrade. If the daemon was started with --db, or with OPENRIG_DB (legacy RIGGED_DB) set differently from this shell, replace the DB= line below with that actual database path. Pass the same --db "$DB" to every rig daemon start, including recovery; rig config get db.path reads this shell's configuration, not the running daemon's launch arguments or environment. A failure stops the following commands; fix it before continuing. Back up and install only after rig daemon stop prints the line Daemon stopped: that is the stop outcome that confirms a clean shutdown. Its other success line, “No daemon target recorded; listener refused. Nothing to stop; prior drain not certified.”, means this stop has no recorded target. It does not prove that no daemon exists or that the last one shut down cleanly. A Daemon stopped from rig daemon status does not confirm a clean shutdown either. If the stop failed or had no recorded target, make sure the daemon is running on its original home and database (start it with rig daemon start --db "$DB" if rig daemon status does not say it is running), then run the whole block again. If the stop is still unconfirmed, use daemon-shutdown.json and daemon.log in your OpenRig home, together with rig context get help, to resolve the reported problem before retrying. Do not continue to the backup or install on an unconfirmed stop. A clean stop after restarting confirms only the new daemon’s shutdown, not whether earlier unfinished actions completed. Each stop replaces daemon-shutdown.json; look in daemon.log for earlier shutdown failures.
Three additive migrations apply at the first start: 176_node_statusline_throttle (the status-line throttle setting), 177_workflow_bound_pod (a workflow instance's pod) and 178_seat_pool (the seat pool's storage). Migration 178 applies whether or not you declare a pool. Keep the backup: downgrading the migrated database to 0.7.19 is not established.
openrig plugin is 0.7.20; openrig-core is 0.1.34.Bundled openrig is 0.7.20; openrig-core is 0.1.34.
openrig-user skill says Claude Desktop picks up the team's work with a scheduled task in a fresh session, and the rigs skill can be added to OMP (Oh My Pi). A community contribution.The openrig-user skill says that Claude Desktop picks up the team's work with a scheduled task, in a fresh session, and that Monitor is for a session kept open. The rigs skill can be added to OMP (Oh My Pi), installing the shared skill only.
--evidence-ref, and scope and proof default to the calling rig's project. A community contribution.A workflow instance can be bound to a pod. A workflow packet's Continuation line names --evidence-ref when its step needs a receipt, as rig workflow show does, and rig workflow route and rig workflow resume point to that exact command. Scope and proof default to the calling rig's project. Current-state progress updates carry a timestamp. An empty plan lock written with CRLF line endings is refused. Block notes are kept beside continuations.
One workflow's unreadable mission no longer hides the attention view's other outcomes. A renamed project keeps its rows by its declared former ids, and rows no project claims are counted. Changes to files.allowlist and progress.scan_roots through the daemon's settings API are recorded on the event stream on a best-effort basis after saving; direct rig config set and rig config reset writes are not covered.
rig start can select fresh seats, and a settled Slack question card names the chosen option. A community contribution.An opt-in per-seat throttle limits how often the Claude status line is collected. rig provider bindings report what each seat is configured to use. rig start can select fresh seats. A settled Slack question card names the chosen option, redacted as the question card is.
A model member can declare a seat-pool block, which is stored on its node, and a sizing planner exists for the pool. Nothing arms or dispatches a pool in this release: declaring one changes no seat's behaviour.
rsync. A community contribution.The TUI degrades its control socket instead of ending the launcher, completes commands with leading whitespace, and measures health findings in terminal columns. The CLI aligns node rows by terminal column width, normalizes explicit daemon URLs, and brackets IPv6 targets in rsync operands. Mission and slice reads ask for the proof policy once per request, and slice-term matching runs in one pass; this release doesn't claim a measured speed-up.
rig start can label a Claude seat resumed before its conversation is confirmed, including when Claude reports that no conversation was found. A send can still land on Claude's workspace trust dialog. Check the seat's screen after a restore or start before relying on it.rig send message sent while Claude's AskUserQuestion dialog is open can become the answer to that question. This was observed on Claude Code 2.1.322 and 2.1.323 with both OpenRig 0.7.19 and 0.7.20.Thank you to everyone whose work is in this release: @aryrabelo, @DeryFerd, @Vaishnavi220506, @Kayhusk, @mvschwarz, @cavalcanticampos, @sn1thy, @dajiaohuang and @billbenetti.
Their contributions also improve the daemon's memory-limit reasons, tests, CI on Ubuntu, the custom-view SQL contract, Claude readiness polling, startup probing, and documentation.
rig seat launch <seat> --fresh replaces a seat whose failed start left only an idle shell, without --stop. New and opt-in: the typing guard's queue-only mode keeps ordinary automatic notices out of a seat you type in, a hook bridge delivers messages to a Claude seat between turns instead of typing into its input box, and on Linux with systemd a seat can request a memory limit. Apps can also read the team's questions and answer them through a versioned interface. Many fixes come from people using OpenRig.5 added4 changed3 fixedA permission prompt can reach the existing alert route even when its seat has no open work. A recovery waiting directly on that prompted seat or row no longer hides the prompt. Clears from an earlier occupant do not clear the current occupant's alert.
Restore now reports skipped saved hooks and missing working folders as gaps. Restore's working-folder projection skips a folder confirmed absent; adapter home/state setup is separate. A fresh launch from a spec can still create its declared folder. Recreated seats' notices and receipts agree about which seat launched, and a delivered sequence can close its notice. Closure is best-effort if bookkeeping fails; a later sequence can still owe work.
rig seat launch <seat> --fresh replaces a seat whose failed start left only an idle shell, without --stop, and can reopen a terminal server after a host restart once the old process is confirmed gone. A community contribution.launch --fresh can replace the shell left by a failed start without --stop. Fresh launch can reopen a terminal server after a host restart when the old process is confirmed gone. A missing tmux server is reported as a detached session, with guidance to restore or relaunch the seat; unavailable readings remain uncertain. Duplicate automatic watchdog registration no longer aborts startup.
For a seat you type in, rig seat set-typing-guard <seat> --mode queue-only --reason <text> keeps ordinary automatic notices out of its input. Watchdog reminders can be noted on a matching row or recorded in a new queue row without a nudge; parked-owner reminders remain retained. Interrupted reminders survive daemon restart. Other delivery modes keep their own behavior.
Claude hook delivery is opt-in: rig seat set-hook-bridge <seat> --enabled true --reason <text>, then inspect rig seat hook-bridge <seat>. On supported Claude versions, launch or restore installs the bridge; it takes messages only after this occupant's waiter has armed. Until then, ordinary typed delivery applies. If an already-routed bridge becomes unavailable, fallback follows the existing delivery checks and names why. The help's 20 seconds is not an arrival deadline: queued work, waiter state and the existing delivery checks determine waiting and fallback. A hook handoff is not proof the agent read or acted on it.
Run the upgrade in the environment that owns your daemon, keeping the same OPENRIG_HOME (legacy RIGGED_HOME) if you set it. Stop the daemon, back up its database, install the patch, then restart it. Your agents keep running; do not use rig down for this upgrade. If the daemon was started with --db, or with OPENRIG_DB (legacy RIGGED_DB) set differently from this shell, replace the DB= line below with that actual database path. Pass the same --db "$DB" to every rig daemon start, including recovery; rig config get db.path reads this shell's configuration, not the running daemon's launch arguments or environment. A failure stops the following commands. Fix it before continuing. If stopping failed, use rig daemon status: when it says “Daemon stopped”, continue from DB=; when it says “Daemon PID is absent”, start the daemon with its original home/database settings before retrying the block.
Three additive migrations record the hook bridge, each session's launched folder, and app access. Keep the backup: downgrading the migrated database to 0.7.18 is not established.
openrig plugin is 0.7.19; openrig-core is 0.1.33.Bundled openrig is 0.7.19; openrig-core is 0.1.33.
The installer guidance keeps your conversation unless you choose the operator's pane, and teaches reloading a skill added during the session. The guidance says an answer covers that prompt, rather than permission to answer unrelated prompts later.
rig seat status shows the limit it observed. A community contribution.On Linux with systemd, process_limits: { memory_max: "512M" } requests a per-seat memory limit alongside tasks_max; swap is separate. rig seat status shows the observed limit and counters. A limit that cannot be verified is reported, not silently claimed effective. Recovery may attempt a launch without a process limit, with a warning; that attempt can fail. An uncertain or already-started command is not repeated merely to obtain the limit.
Apps can use the versioned human-answer interface to read questions, submit answers and recover receipts. Credentials identify an app and permitted actors; they are not private per-person logins. Other improvements bind external input to its run, add optional webhook signatures, and pass advisory participant fields through rig_team.
rig seat cancel asks to stop one exact turn and reports what happened; raw Claude and Codex terminal seats don't get it. A community contribution.Capable runners offer rig seat cancel <seat> --run-id <run> --native-turn-id <turn> --request-id <id> for the exact target from rig seat status <seat> --json. Inspect its requested, cancelled, not-this-turn or not-supported result. Neither permission withdrawal nor an accepted cancellation request proves the operation stopped; raw Claude/Codex TUI seats do not gain this runner capability.
exec change can leave generated settings unrecognized by the older daemon. Keep your settings backup and inspect them before restarting an older release. (#3427)rig compact <seat> --state and rig seat held-messages <seat> before deliberately choosing an existing delivery route or level; changing the level does not replay retained messages.OPENRIG_SESSION_NAME='<recorded address>' rig queue list --owned --state pending,in-progress,blocked.Thank you to everyone whose work is in this release: @DeryFerd, @Vaishnavi220506, @Kayhusk, @EZotoff, @Aummadour, @mvschwarz, @Nikhi00718, @cavalcanticampos, @smota, @sn1thy, @dajiaohuang, @billbenetti and @dependabot.
Their contributions also improve terminal column sizing, JSON errors, transcript searches, workflow ownership, account-label display, scope links, cleanup and documentation.
rig queue create --return-route auto, and when the team stops at a prompt that isn't a choice menu, it comes back to your conversation as a task on your next turn. Nothing wakes an idle conversation, arrival isn't guaranteed, and neither of this release's first-user runs met all its criteria. Restore no longer counts a seat it left as a bare shell as running in rig ps, and three opt-in settings can now be chosen per seat: native input, Claude's auto mode and, on Linux with systemd, a task limit. Many fixes come from people using OpenRig.3 added9 changed5 fixedrig queue create --return-route auto gives the conversation you hand work over from an address, so a seat's prompt that isn't a choice dialog comes back to it as a task; it arrives on that conversation's next turn, and arrival isn't guaranteed.Hand a goal over with a return route: rig queue create --return-route auto gives your conversation an address, and every handoff carries it (--return-route self from a session that joined the team). When routed work is held at a seat's prompt, or waits on it, a prompt that isn't a choice dialog is sent to that address as a task. Choice prompts, a prompt already with the gateway, and a route that can't be used (another host's, two that disagree, or a relay that couldn't be made) keep their existing path. The rigs skill teaches this; without --return-route, nothing changes.
rig ps, and the agent responsible for it gets one task per restore listing each cause and the next step.A seat that restore leaves as a bare shell no longer counts as running in rig ps, including a recreated seat, and a seat with a delegation parent or escalation destination gets that owner one queue item per restore listing each cause and the next step. Launching over a live session no longer says "Already running".
Per-seat choices, all opt-in: native input instead of terminal typing (rig seat set-input-route <seat> --route native --reason <text>); Claude's autoMode block (claude_auto_mode, or rig seat set-permissions --auto-mode); and, on Linux with systemd, a requested task limit (process_limits: { tasks_max: N }). A fresh start whose limit isn't seen to take effect is refused; a recovery that can't apply it may try without the limit, with a warning; and an uncertain start is reported and not retried. An export names a stored limit it can't read instead of refusing the rig.
rig up --existing --cwd restores into the folder you name, and Claude seats keep OpenRig's core plugin.Restoring after OpenRig's home moved: rig up --existing --cwd honors --cwd, names a missing saved folder and leaves a folder with a declared project unscaffolded, and Claude seats keep OpenRig's core plugin.
Stop the daemon, back up the database, install 0.7.18, then start the daemon. Your agents keep running through a daemon restart; don't use rig down, which ends them. Each step runs only if the one before it succeeded. The backup goes to a new, dated file, with its -wal file beside it when there is one, so running the block again never overwrites an earlier backup. If you installed OpenRig another way, use that in place of the npm install line. Before running the block or any recovery command, use the same OpenRig home (OPENRIG_HOME, or legacy RIGGED_HOME) as the daemon you are upgrading. If that daemon was started with --db, or with OPENRIG_DB (legacy RIGGED_DB) set differently from this shell, replace the block's DB= line with its actual database path instead of using rig config get db.path, and pass that path with --db to every rig daemon start below, including recovery. The config command reads this shell's configuration, not the running daemon's launch settings. If a step fails, nothing after it runs: fix what it reports, then run the whole block again. If rig daemon stop fails and rig daemon status then says "Daemon stopped", run the block from the DB= line. If it says "Daemon PID is absent", run rig daemon start, then the whole block again. Keep the backup, and its -wal file, until 0.7.18 is running well.
Four migrations run before the daemon listens (two numbered 171, then 172 and 173), each adding columns or indexes. Going back to 0.7.17 after them isn't established; keep the backup.
openrig plugin carries 0.7.18, and openrig-core moves to 0.1.32.The plugins: openrig carries 0.7.18, and openrig-core moves to 0.1.32 because its skills and refocus hook changed.
messaging-the-human says how to close the question an answer settles.Guidance: the kernel operator is instructed to answer a registered person where they wrote, without the terminal user's leave (guidance, not proof an answer is delivered), and messaging-the-human says how to close the ask an answer settles.
rig seat handover --recover reads an interrupted Claude handover more truthfully and accepts a successor alone in its session's current window. A community contribution.rig seat handover --recover reads an interrupted Claude handover more truthfully and accepts a successor alone in its session's current window.
rig down no longer exits 2 over a released session's kept skills, Codex 0.167.1 and 0.167.2 seats show their real activity, and refocus orients from the seat's project when no task is in progress. A community contribution.Also: a later rig down no longer exits 2 over a released session's kept skills (a plain down of a stopped rig still exits 1); Codex 0.167.1 and 0.167.2 seats read their real activity; an adopted, still-running seat's hooks read as its own; and refocus orients from the seat's selected project when no task is in progress.
rig compact on a Pi seat now asks Pi to compact itself, and --state reports the attempt. A community contribution.rig compact on a Pi seat requests Pi's own compaction, and --state reports the attempt.
rig queue tag adds tags to an existing task, rig queue list names a handed-off task's successors, and a wake that failed before any input gets one retry. A community contribution.Queue and workflows: rig queue tag adds tags to an existing row; rig queue list names a handed-off row's successors; a handoff advises on a mission or slice tag that resolves to nothing; a hold blocked on a custom-ID row records its wake; a wake that failed before any input gets one retry on the same seat; and rig workflow route adopts a second handoff to the same target.
Also: runner reports stay bound to their run, recent transitions honour a limit up to 1,000, and the stuck sweep makes fewer reads. This release makes no speed claim.
TMUX_TMPDIR is named in the attach command, a herdr-app cell at herdr's default hidden size no longer resizes its seat, and rig acp run's log keeps multibyte characters whole. A community contribution.Terminals and assistants: CLI terminal reads send the configured terminal bearer; a custom TMUX_TMPDIR is named in the attach command; with tmux 3.2 or later, a herdr-app cell at herdr's default hidden size (200x120) no longer resizes the seat; tmux credential cleanup can't fall back to a server-wide value; rig acp run's log keeps multibyte characters split across reads; and network assistants learn the retry rule and pickup timing.
.rigged-checkpoint-<session>-<digest>.md; anything that reads .rigged-checkpoint.md won't find it.Each restored seat's checkpoint is its own file, .rigged-checkpoint-<session>-<digest>.md; anything that reads .rigged-checkpoint.md won't find it.
rigs during a running Claude Code session: If installation creates a top-level skills directory that was absent when the session started, Claude Code's documentation says to use /reload-skills (available from Claude Code 2.1.152) to discover it. Starting a new session is an alternative. Neither recovery was exercised in this release's first-user test.--cwd restore, rig down doesn't clean OpenRig's Claude settings, hooks and collector from that folder. When a full restore recreates a seat, its result can still list it as not started, and the attempt's record can miss its actual outcome.human@host asks no one; that holds only for an unregistered one.config.toml or hooks.json is a symbolic link, OpenRig records its hook trust under the link's target, as in 0.7.14. A change to key it by the file's own name isn't in this release, and such a seat may be asked to review hooks again. With a linked config.toml, OpenRig can also miss a setting stored for it that turns OpenRig's hooks off, and set those hooks up anyway when it moves them into hooks.json. (#3078)rig.yaml contains claude_auto_mode or process_limits. Upgrade the receiving install to 0.7.18, or remove those fields from a copy only if you intend to share without those per-seat settings.rig down's exit 2 and Codex 0.167.1 and 0.167.2's unknown (#3185, #3303, #3182).Thank you to everyone whose work is in this release: @Aummadour, @billbenetti, @cavalcanticampos, @dajiaohuang, @DeryFerd, @EZotoff, @oodadoudou, @pr0xc3nt4ur1, @shravansumanthanan, @smota, @Temikus and @w4rc0n.
Thanks also for the reports behind these fixes: @billbenetti (#2725, #3218, #3316), @Kayhusk (#3320), @parkingmeter (#1933), @z4cc (#313) and @ziadeh (#2487).
rig down releases a session OpenRig adopted rather than started, leaving it running instead of failing. OpenRig's installer, kernel operator and rigs skill are also now told to bring a team's questions into the Claude Code conversation you started it from, when you stay in that conversation; in this release's first trial, it didn't complete that journey. There are fixes across delivery, workflows and connected agents, including contributions from people using OpenRig.2 added12 changed8 fixedpartially_restored.Restore says when it didn't restore everything: a seat skipped because it isn't in the snapshot's roster, or left attached, makes the rig read partially_restored. An adopted seat keeps a usable recorded session name; otherwise restore names the old and new address.
rig down now releases a tmux session OpenRig adopted rather than started, leaving it running, instead of failing.rig down releases a claimed tmux session (self-attached inside tmux, adopted, bound or reconciled) instead of failing: it prints Released, not stopped: <session>, leaves the session running and exits 0. An external CLI attachment is unchanged on down, and rig remove of one no longer fails with 409 kill_failed.
When you stay in your own Claude Code conversation, the installer is instructed to hand your goal to the team as a task from it. The kernel operator is instructed to bring a seat's prompt there with its choices word for word, and to offer a seat's terminal only if you choose to watch or answer in the pane. If you chose the panes, nothing changes.
Draft-aware delivery reads an empty input box under Claude Code 2.1.322 or 2.1.323's wide paste hint and auto-update failure as empty (checked with Claude Code 2.1.322 and 2.1.323 in a test run).
Stop the daemon, back up the database, install 0.7.17, then start the daemon. Your agents keep running through a daemon restart; don't use rig down, which ends them. Each step runs only if the one before it succeeded. The backup goes to a new, dated file, with its -wal file beside it when there is one, so running the block again never overwrites an earlier backup. If you installed OpenRig another way, use that in place of the npm install line. If a step fails, nothing after it runs: fix what it reports, then run the whole block again. If rig daemon stop fails and rig daemon status then says "Daemon stopped", run the block from the DB= line. If it says "Daemon PID is absent", run rig daemon start, then the whole block again. Keep the backup, and its -wal file, until 0.7.17 is running well.
Two migrations run before the daemon listens (168 and 170). Going back to 0.7.16 after them isn't established; keep the backup.
openrig plugin carries 0.7.17; openrig-core stays 0.1.31.The plugins: openrig carries 0.7.17; openrig-core stays 0.1.31.
Also: the launch plan says a team seat with no permission policy gets the team default, and every seat's startup text is smaller, with the installing procedure loaded from the rigs skill when needed.
Two opt-in daemon routes: resume one banked original conversation, and hold one rig's automatic delivery during maintenance until an explicit end. See docs/reference/daemon-http-clients.md.
A restore that can't confirm it stopped a blank session keeps that session's record and binding, reads attention_required, and says the session may still be running: check it with rig capture and stop it with rig seat stop.
Claude resume: a pane that moves during resume-token capture leaves the token unknown.
rig workflow revise can drop an obsolete required order from unstarted work, and a replayed route adoption without --packet writes nothing.Workflows: rig workflow revise can drop an obsolete required order from unstarted work, and a replayed route adoption without --packet writes nothing.
Three daemon read paths changed: the stuck sweep yields between windows of undelivered work, Codex thread lookup reuses new log rows on warm reads, with full cold and fallback reads, and a work node's SPEC.md is reread only when its file changed, within a two-second settle window, or when that can't be read. This release makes no speed claim.
disableSideloadFlags now say what to do.Clearer refusals at a model picker (close it with Esc) and under managed disableSideloadFlags.
awaiting-pickup: instead of unroutable:, because it waits until the member reads its queue.Work for a hosted member with no wake trigger reads awaiting-pickup: instead of unroutable:: it waits in its queue until the member reads it.
rig_queue_show labels who acted, the TUI can read a remote rig's seats, Pi and OMP runner panes name their model, and rig bundle check reports invalid configuration names. A community contribution.Assistants and tools: an external connection's rig_queue_show labels who acted apart from the reader; the TUI can read a remote rig's seats with ui.topology_remote_seats (off by default); a Pi or OMP runner pane names its model and state; rig bundle check reports schema-invalid configuration names; and the text output of rig network status hides a hub URL's query (--json is unchanged).
Claude's question dialog, in the several-question, editor-shortcut and preview layouts OpenRig recognizes, refuses an ordinary send (shown by its tests; not yet checked with Claude Code itself).
A resumed Codex conversation isn't told to sign in again because of an old token-refresh failure that newer output follows.
rig provider switch --token-file keeps the Claude token step for pane readings it used to drop.rig provider switch --token-file keeps the Claude token step for pane readings it used to drop, and after a failed correction only for a session proven to be this seat's.
Also: account-rig erasure, OpenShell recreation after a socket change, and handoff reconciliation hints.
undeliverable:no-registered-human.An empty person registry reads undeliverable:no-registered-human.
rig send --dangerously-interact without --type-without-checking now stops on a person's text under Claude Code's wide footer row; its own tests show this, and it isn't yet checked with Claude Code itself.rig send --dangerously-interact without --type-without-checking stops on a person's text under Claude Code's wide footer row (shown by its tests; not yet checked with Claude Code itself).
restore-check's recovery plan still counts seats restore leaves alone, and a seat left as a bare shell after "Harness not started" still counts as running in rig ps.rig down: a second down while a released session runs reports its kept skills as an error and exits 2. After rig down --delete, a later down of another rig in the same folder may remove OpenRig's shared guidance there.unknown when its resume header or update notice names that version.rig scope and rig proof can't be limited to a rig's own project. (#3026)rig queue list: the median time was about 22% (about 90 ms) slower than 0.7.16 in a matched test on a pre-release build. The cause wasn't found, and the final build wasn't re-measured.config.toml or hooks.json is a symbolic link, OpenRig records its hook trust under the link's target, as in 0.7.14. A change to key it by the file's own name isn't in this release, and such a seat may be asked to review hooks again. With a linked config.toml, OpenRig can also miss a setting stored for it that turns OpenRig's hooks off, and set those hooks up anyway when it moves them into hooks.json. (#3078)rig seat handover --recover adopts the running successor of a failed Claude handover, and a message to an agent that doesn't exist is kept and reported as "Recorded, not delivered" instead of only failing. There are fixes across delivery, workflows and connected agents, many from people using OpenRig.12 added18 changed12 fixedrigs skill now tells the coding agent you're working with to bring it to you as a question, instead of answering it itself.The rigs skill tells a person's own agent to bring a seat's permission, trust or hook-review prompt to them as a question by default, instead of answering it itself. It tells the agent to show the prompt, capture the seat again before sending, and send the digit or letter the screen shows for their choice, never their words, and to ask when the screen shows no key. A yes to one prompt isn't a yes to the next, a "don't ask again" choice is picked only when they picked it, and an instruction they gave outright lets the agent decide that kind of prompt within what they said. The agent is told never to record its own answer as the person's. The getting-started guide names each authentication, trust, hook-review or permission prompt as the person's to answer before that seat gets work.
standard or open policy for a seat now keeps the team's default allowances, so it no longer brings more permission prompts than choosing nothing.A seat whose effective policy is the built-in standard or open keeps the team launch default, as a seat with no policy does. Any recorded policy used to turn the team default off, so choosing standard brought more prompts than choosing nothing: ordinary rig reads, git reads and tests all asked. The member's policy decides over the rig's. An explicit seat permission choice, a Codex envelope or named Codex profile, a Claude seat's own rules, none, locked, custom, auto and yolo, and kernel, Pi and terminal seats keep their meaning. This keeps the team default; it doesn't apply the whole policy, and a seat that keeps it can still ask before the project's own commands (see Known limits). The Claude team default includes ask rules for lifecycle commands such as rig up and rig down. The Codex team default does not add equivalent per-command ask rules; native settings and explicit permission choices still apply. The launch plan now says whether a seat gets the team default's allowances, instead of a bare launch_posture=floor. OpenRig can't see rules applied in project or user native settings files; those seats keep the team default alongside them, and keeping it doesn't loosen those rules.
claude-compaction-restore, current when a seat comes back with rig up --existing; copies you edited stay as they are. It's tested at the unit and source level.An upgrade keeps OpenRig's packaged recovery skills. After rig down, an upgrade and rig up --existing, a seat's own unedited copies of OpenRig's packaged skills, such as claude-compaction-restore and seat-continuity-and-handover, come back at the current packaged contents. Before, a packaged skill whose bytes changed in the upgrade wasn't put back: the restore named it missing, and the seat came back without it. When the upgrade came before rig down, the copy used to stay at the earlier version; it's brought up to date now. An edited copy (its bytes or a file mode), a file OpenRig didn't write and a copy recorded from another source stay as they are, and the restore names each. A skill the seat's profile selects itself keeps its own restore policy. Other plugins' skills aren't changed. It's tested at the unit and source level; see Known limits.
rig seat handover <seat> --recover adopts the running successor of a failed handover for a Claude seat, instead of leaving the seat with its old agent; it hasn't yet been tried on a real failed handover. A community contribution.rig seat handover <seat> --recover adopts the running successor of a failed handover, instead of leaving the seat with its retiring occupant. Recovery is for Claude seats: it proves the successor's process and identity from the failed attempt's record, then records the conversation that process holds now as the seat's resume target. When that conversation can't be read, the conversation its launch named is recorded instead, with a warning that a /clear before the recovery would make it stale. Four results are new, all only on --recover: no_handover_to_recover, handover_attempt_superseded, recover_runtime_unsupported and successor_identity_unproven. Recovery from a real failed handover wasn't checked when this release was tested.
A send or row to a missing destination is kept and says so. A queue row whose rig or seat is absent keeps its exact address, with a destination advisory (destination_status, or unmatched_destination_seat for a missing seat in a known rig): missing, archived, hosted or unknown. A direct send that finds no receiver keeps the message in the sender's outbox; the CLI prints "Recorded, not delivered", --verify exits nonzero, and rig walk stops there. No forwarding, retry or reassignment is added.
Stop the daemon, back up the database, install 0.7.16, then start the daemon. Your agents keep running through a daemon restart; don't use rig down, which ends them.
Two migrations run before the daemon listens. Migration 167 adds seat_handover_attempts (for rig seat handover --recover), and migration 169 adds seat_runtime_home_records (for rig seat record-runtime-home). Going back to 0.7.15 after them isn't established; keep the backup.
openrig plugin carries the CLI's version, 0.7.16, and openrig-core moves to 0.1.31.The plugins: the openrig plugin carries the CLI's version, 0.7.16. openrig-core moves to 0.1.31, because its skills changed.
rigs skill tells the agent you're working with to run the rig seat continue command the seat shows, so it gets its startup instructions in the same conversation.After the person answers a prompt a seat stopped at while starting, the rigs skill tells their agent to check rig ps --nodes again and, when it shows Startup details (<seat>) ending in rig seat continue <seat>, to run that command, which delivers the seat's startup instructions in the same conversation. When the person's choice is Escape, it tells the agent to send the key itself (rig send <seat>@<rig> $'\e' --dangerously-interact --reason "…"), which is sent once, with no Enter added, and to repeat the prompt check while it waits for a seat's result.
A person's assistant on an external grant sees the team by the name the person knows it by, and its members by their roster label. The team's name comes from the rig's network roster or, when there's none, from the roster that declares the rig. A label over 200 characters, or with control characters, line or paragraph separators, bidirectional embeddings, overrides or isolates, certain invisible formatting characters, or angle brackets, is set aside with label_problem; the member and the roster still load.
rigs skill teaches the agent that installed OpenRig for you to keep its own return address, so the team's questions come back to it.The rigs skill's setup path teaches an assistant that installed OpenRig for a person, and that they keep talking to, to keep its own return address, so the team's questions come back to it and it answers them with the person's exact words.
OpenRig's MCP instructions tell the assistant to list the team and ask which assistant a person means when more than one could be, instead of picking one from its own history. rig mcp serve --front-door says a request is picked up on its own only by a running team seat.
rig seat runtime-home shows which runtime home a seat's next launch uses, and rig seat record-runtime-home records the one its existing conversation lives under, with no relaunch.rig seat runtime-home <seat> shows the runtime home (Claude Code's config directory or a Codex home) the seat's next launch uses. rig seat record-runtime-home <seat> --home <abs> --expect-occupant <gen> --expect-original <id> --expect-revision <n> --reason "<why>" records the one its existing conversation lives under, with no successor, relaunch or manual database edit. Recording sets the home used by the seat's next launch. A later explicit home change is retained; restoring a selected conversation shows and applies the home selected for that original. A stale expectation returns runtime_home_conflict and applies nothing. The result reports whether the home is present and whether the original conversation is found there, and reports the running process as unverified.
When managed restore or preparation fails for a seat with no independent delegation parent, the recovery notice goes to the receiving agent or kernel operator, then to a registered person. Existing notices keep their ids and aren't rerouted.
rig seat set-runtime switch settles as abandoned, so sends to the seat go through again; it hasn't yet been tried across a real reboot. A community contribution.After a host reboot or a tmux server restart, a rig seat set-runtime switch that was still before commit settles as abandoned once its deadline has passed, its staging session is gone and no observed pane shows the staged successor. Sends to the seat then go through again. Any sign of the successor keeps it pending, and the deadline is an estimate, not proof that nothing is in flight. It settles only at daemon start or on rig seat set-runtime <seat> --recover, and it wasn't tried across a real host reboot or tmux server restart.
rig compact that stops at a prompt now waits and sends /compact once the seat is idle, and --skip-map can compact a seat at Claude Code's "Context limit reached" screen.A manual rig compact whose preparation stops at a prompt returns to its bounded wait instead of disarming, and sends /compact once the seat is idle. Nothing is answered for the person, and a typed draft still stops the attempt. rig compact <seat> --state shows the prompt; an expiry while it still shows names preparation_needs_input. A manual compaction also no longer needs a context-usage reading, so rig compact <seat> --skip-map now compacts a seat at Claude Code's "Context limit reached" screen even with no usage reading (checked once, in Claude Code's default mode).
An attached or adopted Claude seat can capture its current conversation's resume token from process metadata when no seat-named status-line file exists, and Claude's conversation title no longer has to equal the seat name. A missing bound pane or an ambiguous session leaves the token unknown. Plain external-CLI attachment without a tmux pane stays unverified, and restore still relaunches an adopted seat under a derived session name.
rig launch <rig> --seats <seat> no longer refuses because another seat is paused or being typed in, and it takes a seat's session name.rig launch <rig> --seats <seat> no longer refuses because another seat in the rig is paused or being typed in; the other seats keep their pane, draft, guard mode and binding, and a paused target still refuses. rig launch also takes a seat's session name, including an adopted seat's raw tmux name or its <name>@<rig> form. A name matching several seats isn't launched. The single-seat form returns seat_ambiguous with candidates; under --seats it is listed in ambiguousSeats while other unambiguous selections proceed normally. If every selection is ambiguous, the reply is seat_ambiguous.
For the default http://127.0.0.1:<port> target, the TUI's daemon-only start (from its crash cart or start-and-return page) passes only the port, so it honours the daemon's configured bind, or its default loopback-plus-Tailscale bind when none is configured, instead of loopback alone. It no longer promises a kernel that isn't listed.
--summary or --evidence-ref now stores them, instead of refusing the whole update.An otherwise-admitted state-changing park or close stores its supplied non-blank --summary or --evidence-ref, instead of refusing the whole update with summary_evidence_not_persistable; note-append and re-estimate rules remain unchanged. A blank value isn't stored, and the receipt names any requested field the stored row doesn't carry.
rig ps summary lines now name the daemon they came from. A community contribution.rig ps ends each summary line with the daemon it answered from (· daemon <address>); JSON output is unchanged.
A watchdog records the request's session header as its registering session when there is one. That's attribution, not an authenticated identity.
The queue's waiting reads reuse their prepared statements per connection. Results don't change, and this isn't claimed to fix a daemon freeze.
A draft typed while a conditional Enter's own checks wait is no longer submitted with OpenRig's text: the guard refuses with draft_input_changed, sends no key and leaves the input as it is. The input read and the later key write are still separate steps; see Known limits.
A send to a busy Codex seat (seen with 0.167.1 and 0.167.2) whose input is empty is delivered once, instead of being pasted and left in the input, neither delivered nor held. OpenRig now reads Codex's "tab to queue" hint row, and a person's draft under that hint is still held. That's shown by the change's own tests; it wasn't checked with the real programs when this release was tested.
human@host no longer suggests replying there, because that address names no person.A notice from human@host no longer hints at replying there: the hint says that address names no person.
rig edge add no longer says restore refuses a cycle, and a ceremony episode keeps its name when what counts as coordination changes. A community contribution.rig edge add's cycle refusal no longer says restore refuses a cycle, and an open resumed ceremony episode keeps its name when what counts as coordination changes.
rig workflow list filters by mission, project, rig, workflow, slice or tag. A community contribution.rig workflow list filters by mission, project, rig, workflow, slice or tag; repeated tags combine with AND.
rig workflow route to the same target adopts the handoff's successor instead of writing a new task.When a queue handoff closed a workflow's frontier packet, rig workflow route to the same target adopts the queue's single handoff successor, writing nothing to the queue, and rig workflow project records the agent's exit on the closed packet without changing its queue history. A route that can't adopt returns 409. A write-free replay needs the same original --packet; a canceled packet isn't treated as terminal; a lone handed-off graph packet projected as waiting still reports the graph as active; and route doesn't mark a parent workflow recovered.
rig ps, rig send, rig capture and the queue can name a registered remote host, from an owner connection. A community contribution.MCP tools whose CLI verb takes --host accept a host for a registered http host: rig_ps, rig_send, rig_capture, and the queue create, show, list, update and handoff tools. Only an owner connection can name a remote host; a granted connection is refused, and an ssh host gets a teaching error.
rig mcp token rotate <name> gives an MCP grant a fresh token in place; clients must switch at once. A community contribution.rig mcp token rotate <name> gives a file-backed MCP grant a fresh token in place. Clients must switch: the old token stops matching at the next authentication, and there's no grace window.
codex_permissions; enforcement hasn't yet been checked with a real Codex seat. A community contribution.Per-seat Codex filesystem policies: codex_permissions has an opt-in filesystem form with native read, write and deny rules for that seat's own invocation, carried through managed launch, resume, restore and rebuild. A concrete conflict in the seat's effective Codex configuration stops the launch and names a remedy; a configuration that can't be read keeps the authored invocation and reports "policy not verified". The configuration read and the start aren't atomic. Codex enforces the rules. A seat's resume command in rig ps and node detail carries the policy's flags, but not yet its model and effort. When this release was tested, enforcement wasn't checked with a real Codex seat, and neither were a manual resume with that command nor two seats keeping their policies apart.
The TUI's Projects view can show missions and slices across every registered host, each with a status (ok, stale, unreachable, auth-failed, unsupported or unsupported-transport). A failed read shows that host's last-known rows marked stale; that cache is in memory, so after a daemon restart a failed host shows no rows.
rig ps --nodes --json --fields activityState returns each seat's activity, and daemon.seat_poll_interval_ms says a change applies after a daemon restart. A community contribution.rig ps --nodes --json --fields activityState returns each seat's activity object, and daemon.seat_poll_interval_ms reports that a change applies after a daemon restart.
scripts/portability-report.mjs --strict exits 1 when the report finds matches. A community contribution.scripts/portability-report.mjs --strict exits 1 when the report finds matches; without it nothing changes.
openrig plugin's README gives GitHub Copilot CLI's install and seat steps.The openrig plugin's README gives GitHub Copilot CLI's install and seat steps; Copilot installs the plugin from the repository's main branch, unlike the other hosts' release installs.
rigs skill says how to follow a task handed on to another host, and that rig queue list shows 100 tasks by default.The rigs skill says how to follow a task handed on to another host (rig queue show <id> --host <host>), and that rig queue list shows 100 rows by default.
Docs and tests: #3197, #3201 and #3203.
rig context get reference/product-journey-sdlc.md resolves. A community contribution.rig context get reference/product-journey-sdlc.md resolves.
standard or open policy now launch with the team default's allowances; before, any recorded policy turned the team default off.standard and open seats keep the team default. A seat whose effective policy is the built-in standard or open now launches with the team default's allowances, as a seat with no policy does. Before, any recorded policy turned the team default off.
--summary or --evidence-ref is now applied instead of refused.A state-changing park or close carrying --summary or --evidence-ref is applied instead of refused.
rig ps summary lines end with · daemon <address>. A community contribution.rig ps summary lines end with · daemon <address>.
fully_restored while a seat restore didn't start (one claimed or added after its roster or snapshot was recorded), or a seat kept in its external CLI attachment, goes unnamed. A seat restore left as a bare shell after "Harness not started" still counts as running in rig ps. restore-check can also say "Restore tries …" for a self-attached member that restore leaves out; that member isn't restored.rig seat held-messages shows it.rig commands such as rig whoami and rig queue show. A seat with the built-in standard or open policy now keeps the team default; other recorded choices keep their meaning, and the launch plan says whether a seat gets the team default's allowances. A seat that keeps the team default can still ask before the project's own commands, such as running its tests or git add and git commit. Each prompt waits until someone approves it, in the seat or through your own agent, so a team can stall while nobody answers./compact can be sent as part of that text instead of as a command. The seat isn't compacted, but rig compact --state reports the compaction, restore and audit as sent. Make sure the input box is empty before compacting, and check that the seat actually compacted.[Pasted text #1]), and the attempt stops as preparation_incomplete.rig down had already removed the first file, the copy is kept and named rather than followed. If saving a skill's new record fails, files whose source changed are named missing, as before, and a later restore tries again.config.toml or hooks.json is a symbolic link, OpenRig records its hook trust under the link's target, as in 0.7.14. A change to key it by the file's own name isn't in this release, and such a seat may be asked to review hooks again. With a linked config.toml, OpenRig can also miss a setting stored for it that turns OpenRig's hooks off, and set those hooks up anyway when it moves them into hooks.json. (#3078)Thank you to everyone whose work is in this release: @DeryFerd, @billbenetti, @EZotoff, @Kayhusk, @dajiaohuang, @cavalcanticampos and @w4rc0n.
Thanks also for the reports behind these fixes: @billbenetti (#2985, #2728), @KM199 (#3103) and @Kayhusk (#3193).
OpenRig renews its own changed Codex hook approvals. When a new openrig-core changes one of OpenRig's six Codex hook declarations, the launch that installs it renews those approvals from OpenRig's own record instead of stopping at Codex's hook review. This runs at daemon start in the daemon's Codex home, when hooks are on and a Codex seat selected openrig-core, and at each such seat's launch in its own Codex home. It renews only when the installed plugin matches the shipped one and every current approval matches OpenRig's saved record. A person's own choice in any layer OpenRig reads keeps the review; profile files passed with -p aren't read. No real Codex seat ran in its tests.
rig daemon start shows its progress about every 20 seconds.A slow start: rig daemon start now gives the daemon child it started a 180-second readiness budget, instead of 20 s, and prints "Waiting for daemon child <PID> readiness (…s elapsed; 180s limit)" about every 20 s. A quick start still returns at once, and a child that exits ends the wait. Let the wait finish. The budget bounds scheduling: synchronous work can run past the budget. A child still not ready when that budget is exhausted is reported pending as before: "Daemon startup is pending after the readiness deadline". Readiness remains unconfirmed; the child keeps running and the startup reservation is retained. Inspect the address and log it prints, and don't start it again until its restart steps are done: stop the named child PID and wait for it to exit, inspect daemon-start.lock, and only when both recorded launcher and child PIDs are absent, archive the lock and run rig daemon start. rig daemon stop cannot stop that unpublished child.
For an eligible seat in a rig with pods that restore starts in its own pane, a resume token with no saved startup context now lets restore try the saved conversation with an empty plan. It records the startup_context gap. A successful resume with that gap makes the rig read partially_restored; a missing conversation or failed start keeps its own status. No saved startup plan is replayed or saved startup-context row written; the adapter's ordinary bootstrap and existing skill restoration still run. An external CLI attachment is kept, no-token seats still wait for a decision, and legacy seats are unchanged. restore-check reads an eligible seat with a token yellow; for one with no saved context and no eligible token, it points to the read-only rig up <rig> --existing --plan preview.
rig slack setup --unfurl on to turn them back on. A community contribution.Slack previews: posts to a person no longer show link and media previews by default. An existing install stops showing them after its next connector rewire or daemon restart. Run rig slack setup --unfurl on to turn them back on; earlier previews aren't removed.
Stop the daemon, back up the database, install 0.7.15, then start the daemon. Your agents keep running through a daemon restart; don't use rig down, which ends them.
One migration runs before the daemon listens. Migration 165 adds four columns and two indexes to activity_grants, binding a runner's activity credentials to its actual launch. Going back to 0.7.14 after it isn't established; keep the backup.
rig daemon start shows its progress about every 20 seconds.A slow start: rig daemon start now gives the daemon child it started a 180-second readiness budget, instead of 20 s, and prints "Waiting for daemon child <PID> readiness (…s elapsed; 180s limit)" about every 20 s. A quick start still returns at once, and a child that exits ends the wait. Let the wait finish. The budget bounds scheduling: synchronous work can run past the budget. A child still not ready when that budget is exhausted is reported pending as before: "Daemon startup is pending after the readiness deadline". Readiness remains unconfirmed; the child keeps running and the startup reservation is retained. Inspect the address and log it prints, and don't start it again until its restart steps are done: stop the named child PID and wait for it to exit, inspect daemon-start.lock, and only when both recorded launcher and child PIDs are absent, archive the lock and run rig daemon start. rig daemon stop cannot stop that unpublished child.
openrig plugin carries the CLI's version, 0.7.15, and openrig-core moves to 0.1.30.The plugins: the openrig plugin carries the CLI's version, 0.7.15. openrig-core moves to 0.1.30, because its skills changed.
The slow-operation log adds three queue sites and three selected steps of an ordinary nodes read: nodes.inventory.build, nodes.inventory.activity_sync and nodes.inventory.terminal_work. These nodes records are written after a slow call completes; a missing record does not rule out unfinished work or work outside those steps. This changes no behaviour.
The wake ladder reads its prompt-refusal candidates from the open rows, instead of scanning the whole transition log on every tick. It returns the same rows. This removes one repeated scan; it isn't a measured fix for every stall.
The execution view's candidate read steps through the queue in windows and, on its cooperative path, yields to the event loop between them, so a mission filter that matches few rows no longer holds the loop for a whole history walk. One large row, or the history count, can still take time. Inside a transaction, on a memory or temporary database, or on a non-WAL file, the candidate read keeps its synchronous fallback.
restore-check and a full restore warn when a snapshot's named Codex profile has since been cleared from the seat; restore uses the current one. A community contribution.restore-check and a full restore warn when a snapshot recorded a named Codex profile that the current seat has cleared; restore uses the current selection. Nothing is refused.
rig seat restart --keep-conversation can't restart a seat whose tmux session is gone, it gives you the rig launch command that brings it back. A community contribution.When rig seat restart --keep-conversation refuses a recoverable seat whose tmux session is gone, it names the way back: rig launch '<rig>' '<node>' (add --plan to preview).
A context-usage watchdog reads its own seat's current occupant, not an archived generation's. A context watch whose seat can't be pinned, or has no recorded tenure, skips that tick with its reason instead of ending.
rig slack setup --unfurl on|off, and they're off by default. A community contribution.Slack previews on posts to a person are a setting: rig slack setup --unfurl on|off, and rig slack status shows which is in effect from the connector's next rewire. The default is off; see "Behaviour that changed".
Right before an automatic paste into a seat with the draft-aware guard, OpenRig re-checks: it holds the paste after recent client activity, re-reads the input after a short wait, and holds when tmux can't be read. The message is held and retained; rig seat held-messages shows it. An ordinary guarded paste waits about 300 ms longer. This narrows #3076's window; it doesn't close it. This is shown by the change's own tests; it's not yet checked with the real programs on this release.
A terminal seat whose tmux session runs under another name gets the exact text of a send, and a queued wake no longer types OpenRig's From/To envelope into its shell.
When the arrival note is recorded before the answer is routed, an answer to a reply in an older Slack thread posts back into that thread, not as a new top-level message; see Known limits for when it posts top-level. Only the conversation's newest thread answers a parked question, as before.
rig_team for its team by display name is told to leave rigId out, and a hosted member with no wake trigger is told when it picks up work.An assistant that passes its team's display name as rig_team's rigId is told to leave rigId out to see its own team, and a hosted member with no wake trigger is told it picks up work when its person asks, or on a schedule they set.
An external grant that works its own inbox is told only about the tools it's offered.
openrig plugin needs Claude Code 2.1.119 or later; on earlier versions, with its seat setting left blank, the skill loads but the agent has no OpenRig tools.The openrig plugin needs Claude Code 2.1.119 or later. On earlier versions, with the plugin's OpenRig seat setting left blank, Claude Code drops its MCP server: the skill loads, but the agent has no OpenRig tools.
The TUI keeps connected assistants' check-in lines when a rig's seat inventory can't be read, with the error beside them.
Runner activity grants are bound to the actual launch, and the replacement's activation revokes its predecessor's grant. A grant made before this release stays tied to the seat's occupant until then. A grant that can't be bound never blocks a launch or restore. This attributes reports to the run OpenRig selected; it doesn't prove a runner's identity.
A Codex seat's resume command in the node inventory and detail panel is built from what a real resume uses, with a label when the recorded launch differs. Workspace options that aren't recorded per launch can still read as a match.
Terminal views find tmux again when its cached path is gone, instead of tiles exiting 126 until a daemon restart.
SKILL.md with Windows (CRLF) line endings is now found and reaches the seat's managed plugins. A community contribution.A SKILL.md with CRLF line endings is discovered and reaches a seat's managed plugin loadout.
rig context work-install --mission warns when the mission doesn't exist.rig context work-install --mission <missing> warns that the mission doesn't exist, and missionRoot reads null; the exit code is unchanged.
OpenRig's own transitions (daemon@kernel and system:*) no longer count toward ceremony amplification, and the post-restore audit message matches the restore skill.
partially_restored when it works, instead of stopping as not restorable.Restore of an eligible seat with a resume token and no startup context now tries its saved conversation, and a successful resume reads partially_restored. It used to stop with "Harness not started: no startup context" and read the rig not restorable.
fully_restored while a seat restore didn't start (one claimed or added after its roster or snapshot was recorded), or a seat kept in its external CLI attachment, goes unnamed. A seat restore left as a bare shell after "Harness not started" still counts as running in rig ps. restore-check can also say "Restore tries …" for a self-attached member that restore leaves out; that member isn't restored.rig seat held-messages shows it. On 0.7.14, the clipped footer at 80x24 didn't do this.rig seat held-messages. A deliberate send reports failed, and its text stays in the seat's input box. This was seen on 0.7.12 and 0.7.14, and there's no fix for it in this release.rig commands such as rig whoami and rig queue show. Choosing a built-in policy such as standard when your team is set up doesn't change that at launch: the plan shows launch_posture=floor. Each prompt waits until someone approves it, in the seat or through your own agent, so a team can stall while nobody answers.config.toml or hooks.json is a symbolic link, OpenRig records its hook trust under the link's target, as in 0.7.14. A change to key it by the file's own name isn't in this release, and such a seat may be asked to review hooks again. With a linked config.toml, OpenRig can also miss a setting stored for it that turns OpenRig's hooks off, and set those hooks up anyway when it moves them into hooks.json. (#3078)Thank you to everyone whose work is in this release: @dajiaohuang, @Kayhusk, @billbenetti, @Aummadour, @smota, @vanducng, @zichen0116 and @Nikhila-Dhevi-R.
Thanks also for the reports behind these fixes: @Aummadour (#2989), @vanducng (#3128), @Alejandro-SJ (#2990), @billbenetti (#2976) and @Kayhusk (#3131).
Before your first team starts, setup and rig doctor say which logins it needs, and the operator lets you choose if one is missing.
rig queue list filters exactly by --tag, --mission and --slice, and several tags must all match. Thanks to @Aummadour.
rig terminal open --provider herdr reuses a live Herdr workspace, and Rex tiles show each agent and what it's doing. Thanks to @Kayhusk and @Tankard.
Stop the daemon, back up the database, install 0.7.14, then start the daemon. Your agents keep running through a daemon restart; don't use rig down, which ends them.
Three migrations run before the daemon listens. Migration 163 adds the network_installations table and four columns to network_bindings. Migrations 164 and 166 build indexes. The author's measurement of 166 took 3.47 s on 28,976 rows, not a real installation's size, so a large database can take longer. The write overhead, and going back to 0.7.13 after these three migrations, aren't established; keep the backup.
rig daemon stop can't stop that child.If the start is reported pending: the CLI waits 20 s, then may print "Daemon startup is pending after the readiness deadline". Readiness remains unconfirmed. It leaves the child running and the startup reservation retained, without recording the daemon. Inspect the address and log it prints. The child may later answer, and status can then report it running, but that does not finish this start's recording or release its reservation. Don't start it again until the steps below are done. The message's restart steps stop that child and clear the reservation: stop the named child PID and wait for it to exit, inspect daemon-start.lock, and only when both recorded launcher and child PIDs are absent, archive the lock and run rig daemon start. On a large database the new start can be reported pending again; reaching a recorded start this way hasn't been tested. rig daemon stop cannot stop that unpublished child.
openrig plugin now carries the CLI's version, 0.7.14; openrig-core stays 0.1.29.The openrig plugin carries the CLI's version, 0.7.14; openrig-core stays 0.1.29.
rig ps --nodes, the TUI and busy agents now take turns, one step at a time, instead of arriving as one burst.Node reads (rig ps --nodes, the TUI, several agents at once) take turns, one step at a time. In the change's own tests, health checks and queue reads were no longer held behind a whole burst. A comparison under load didn't attribute a gain to this change: health checks and queue reads were faster for the release as a whole, rig ps --nodes showed no change, and sends were inconclusive.
Bounded queue listings over retained history use indexes for their ordering and for live and archived owner notifications.
Codex seats re-checked in one readiness poll share process listings where each seat's observation order and the 2 s freshness allow; each keeps its own later observation and its own fallback.
New tmux servers raise their soft open-file limit to 8,192 or the hard limit where they can, keeping an already higher limit. This makes descriptor exhaustion less likely as panes and viewers add up; servers already running keep their limits.
A delegation parent gets one standing notice per child seat for unclaimed rows, not one per row, and held work gets its notice back when its typing hold clears.
An assistant connected over MCP reaches a seat by its plain name, stored under the seat's actual session; an unknown name, or one several seats match, says so.
This is for hosted agents connected through a grant service (experimental); seats OpenRig launches behave as before.
rig ps, rig_team and the TUI.Connected assistants appear, with their check-in, in rig ps, rig_team and the TUI's tables, and their person sees plain tool titles.
A team your agent hands work to knows how to ask you a question, and your answer wakes it. A click on an ask with several questions records that choice; the ask stays open until the last one.
openrig plugin for Claude's plugin directory follows handed-on work and keeps its reply route.A skill-only edition of the openrig plugin, for Claude's plugin directory, follows handed-on work and keeps its reply route.
When the readable person registry holds nobody, rig daemon start and rig gateway participants tell the setup agent to offer registration. The kernel operator is taught to include that offer in its greeting and register only on your yes.
A pairing request's description of itself is quoted wherever it's shown.
/model picker in Claude Code or Codex: it's refused until you close the picker, though this hasn't been checked with the real programs yet.An ordinary send is refused as needing input while an open Claude Code or Codex /model picker is positively identified; nothing is typed. Close the picker before trying to deliver the message. This is shown by the change's own tests; it hasn't yet been checked with the real Claude Code and Codex programs on this release.
rig up --existing, no longer adds OpenRig's guidance to your tracked AGENTS.md on Codex 0.164.0 or later in a rig with pods.On a resume of a Codex seat's own conversation in a rig with pods, including rig up --existing, OpenRig avoids adding missing guidance blocks to the repository's tracked AGENTS.md when Codex is 0.164.0 or later, no guidance file or developer instructions are configured, and the configuration and block sources can be read.
With recovery.auto_drive_provider_prompts on, Claude Code's gated-grants trust check is never answered for you, and the plain trust prompt is answered "Yes".
rig broadcast now shows a held recipient as retained, not sent.rig broadcast prints a held recipient as retained, not sent.
Account erasure also removes OpenRig's own wakes and notices for that account's seats.
A snapshot without sessions returns its invalid_snapshot_data blocker instead of throwing a TypeError; the restore still stops there. A retried no-launch import recovers the original rig by its operation ID.
A rebuilt briefing limits optional superseded recap paths to the newest three by default when one folder covers that history; the current recap and all artifact records remain. Other layouts keep the full list, and output.rebuild_recaps set to 0 lists all paths.
rig compact --state names the input check that stopped a managed compaction, and the TUI graph shows an unknown agent as unknown.rig compact --state names the input check that stopped a managed compaction, and the TUI's graph shows a seat the daemon reads as unknown as unknown.
rig workflow validate warns that harness pins aren't checked. A community contribution.Setting a context-pressure threshold explains when an environment override inverts the pair, and rig workflow validate warns that harness pins aren't checked against seats.
The config browser's entries say each setting's type, choices, range, format and write route.
rig network is experimental and unsupported, and its connector runs even with nothing bound.rig network is experimental, unsupported and at your own risk. Its connector isn't inert: the daemon starts it, sweeps leftover credential folders and schedules local rounds even with no binding.
A grant service can hold back only new work under its fair-use limit, and an app can read what needs a person, and the answers to it, by link.
Node reads: a nodes response is no longer one snapshot across its steps, and in a burst each nodes response comes later: about 43 ms against 21 ms median in a 24-request burst. One very large rig is still one step.
An assistant naming a seat now hears three errors: destination_ambiguous (several seats match; the candidates are listed), destination_not_found (no seat it can reach has that name) and destination_unchecked (the team couldn't be read; try again).
rig roster still prints it. (#3105 isn't in this release)git status. A block whose source can't be read is reported missing.config.toml or hooks.json is a symbolic link, OpenRig records its hook trust under the link's target, as in 0.7.13. A change to key it by the file's own name isn't in this release, and such a seat may be asked to review hooks again. With a linked config.toml, OpenRig can also miss a setting stored for it that turns OpenRig's hooks off, and set those hooks up anyway when it moves them into hooks.json. (#3078)Thank you to everyone whose work is in this release: @Kayhusk, @billbenetti, @cavalcanticampos, @Aummadour, @Tankard, @w4rc0n and @Totopo27.
Thanks also for the reports behind these fixes: @Kayhusk (#3035), @Tankard (#2583), @dajiaohuang (#2529), @billbenetti (#2728), @emilio-gagliardi (#3087), @korallis (#161) and @w4rc0n (#3017).
When a runner agent asks permission, your answer now gets a note back: who decides it, and whether the tool ran. The waiting agent has the decision in about half a second. If what you typed isn't clear, the agent reading it asks you back in your Feed, and your reply decides. Closing a request without deciding permits nothing.
A Claude agent that hits its usage limit now waits for the reset time in Claude's message, for example "resets Nov 10, 4pm", and picks its work back up then. Before, it was retried every few minutes. Thanks to @cavalcanticampos, and to @alperta for the report.
Before your first team starts, the operator tells you which commands it will ask about. It then makes one offer: allow the command families your goal repeats, for this project only. Say no and the team's defaults stay, and the offer isn't repeated.
Setting up OpenRig now offers to register you as the team's person, so your agents' questions have somewhere to go. rig daemon start and the rigs skill both make the offer. Nothing is registered without your yes, and rig gateway human add registers you any time.
OpenRig now keeps your edits to its blocks in AGENTS.md and CLAUDE.md, and removes duplicate copies it recognizes. You can tune its guidance for your project without the next update undoing it.
Back up the database, install 0.7.13, then restart the daemon: rig daemon stop, copy the database file (rig config get db.path prints its path), install OpenRig 0.7.13 using your existing installation method, then rig daemon start. Your agents keep running through a daemon restart; don't use rig down, which ends them. Three migrations run on the daemon's first start: 160 and 161 (for the experimental hub connector below) and 162 (indexes for session reads). Going back to 0.7.12 after them isn't established.
openrig plugin now shares the CLI's version number. If a Codex agent asks for hook approval after the upgrade, approve the hooks to continue.openrig-core is 0.1.29, and the openrig plugin now carries the CLI's version, 0.7.13. A Codex seat can ask for hook approval after the upgrade; see Codex hook review after an upgrade below.
A connected assistant (a hosted member; experimental and off by default) shows whether it's checking in, against a cadence its roster entry states (check_in_minutes, or check_in_on_request: true): checking in, overdue, no check-in yet, on request, or unknown. rig roster show and find print it as a Check-in line. It's a status only: work for a member that isn't checking in still queues and rings as before.
A handover whose successor was launched but not ready says it may be running, instead of "retry": look at the pane first. A handover also cleans up the retiring occupant's own claims and jobs.
rig unclaim, the guidance tells you to end the adopted session's process and tmux session, and a fresh launch reopens the tmux server when it needs to.After rig unclaim, the guidance for an adopted session says to end its process and its tmux session, and rig seat stop's help matches. A fresh launch after rig unclaim ended the server's last session reopens the server.
A handover whose successor was launched into the seat's own tmux session can reclaim a claim left by a removed owner, or by an archived owner whose latest session for that node is superseded, detached or exited; a live owner still refuses. When a seat with no declared route can't be woken, the finding goes to the item's creator, skipping a creator that is a terminal seat.
Speed: the daemon is much more responsive while it reads a project's proof, which now yields between missions; current-session reads for the node inventory use an index; the daemon's recorder and Linux clipboard discovery no longer wait on the main thread.
rig seat clear-attention checks, how compaction notices are routed, and how to start one stopped kernel seat with rig launch. A community contribution.Docs: what rig seat clear-attention's check proves, how managed compaction notices are routed, when the keyboard moves in herdr-app, a complete periodic-reminder registration, and starting one stopped kernel seat with rig launch.
Codex and Claude: completed Codex 0.167 turns read idle, and unsupported versions stay unknown; a usage-limit banner with a curly apostrophe is detected; refocus bounds Codex context and records what it left out; an unreadable Claude status-line cache gets its own reason.
TUI and CLI: instance rows are grouped by rig; for a quiet seat the daemon explicitly reads as unknown, the list, explorer and agent-detail status rows show unknown, not idle (screen motion still shows active, a daemon that serves no activity keeps the screen fallback, and the topology graph is unchanged); Feed times use the configured time zone; the CLI doesn't call a daemon stopped when the shell can't see its process (a sandbox), and says not to start another; ssh options are separated from the target, and the generated attach command quotes the session name.
Scope and workflows: qualified slice reads and proof discovery are scoped by mission; workflow spec reads are recorded across restarts; the spec library skips synthetic alias entries.
rig network connects one local seat to a hub you already have. Experimental and unsupported.rig network connects one local seat to a hub you already have. Experimental, unsupported and at your own risk. The connector uses the MCP SDK, which the installed package already included in 0.7.12.
A runner's answer is confirmed to you. Through 0.7.12 no note came back, and the runner waited for its next read.
A usage-limited Claude seat waits for its stated reset; don't release it by hand on a timer. A failed handover may have left its successor running; look at the pane before retrying.
rig host doctor checks them. A community contribution.Passive SSH host status reads unknown. Routine host listings no longer open a TCP connection to SSH hosts, because repeated probes closed without a handshake can make OpenSSH block real logins for a while. Hosts that used to read reachable now read unknown; rig host doctor still checks actively.
rig package install leaves a guidance file with a damaged OpenRig block alone and lists it, instead of adding a second copy.rig package install leaves a guidance file alone when a recognized managed block has no complete marker pair, lists it under "Left as it is:" and installs the rest, instead of appending a second copy. Launch and restore name such a file (guidance_block_markers_damaged).
rig up --existing can write OpenRig's managed guidance into a tracked AGENTS.md, leaving the worktree modified; check git status after either operation.rig up, rig start and rig context start the daemon without printing the offer either. Register with rig gateway human add. While nobody is registered, a later rig daemon start offers again, even after a no; nothing is registered without your yes. (#3028, #3041 isn't in this release)rig queue create invocations had a median about 10.4% slower than 0.7.12 (about +29 ms). Send latency stayed within the comparison's unchanged range. The cause remains unidentified. The standard idle-memory comparison was inconclusive. A post-hoc analysis of the same six runs found 7.9–8.6 MiB (4.2–5.3%) higher daemon RSS across three matched idle windows, using the median of three arm medians per version. The cause and steady-state behavior remain unestablished.rig up --existing stopped for hook review in each Codex pane; approving the hooks resumed all three conversations. A separate test of the same upgrade on a fresh Codex home, on another host, restored without review. The cause of the difference is not established.&&, ; or a pipe still asks unless every part is allowed, and OpenRig's own lifecycle commands still ask on Claude seats. (#3029)claude -p. If an interactive seat's limit report doesn't carry the message, the seat is retried as before. A reset in the past or more than 8 days away is ignored. (#2917)rig down keeps no stored host or boot identity, doesn't guard against another program running as you, and a daemon restart between losing the tmux server and the stop wasn't exercised. (#2962)agents.operator_session) that runs in an explicit or auto permission mode gets no OpenRig allowance for rig host approve or rig host deny, so Claude's review can refuse them.rig seat stop no longer suggests rig down for an adopted session (#3006). They weren't re-checked for this release unless named here.Thank you to everyone whose work is in this release: @Aummadour, @Kayhusk, @Coder8124, @DeryFerd, @EZotoff, @hgreene624, @cavalcanticampos, @ziadeh, @dajiaohuang and @billbenetti.
Thanks also for the reports behind these fixes: @jamesloosli (#2879), @Kayhusk (#2880 and #2951), @CaptObvious (#2948 and #3016), @alperta (#2917), @ziadeh (#2960) and @todd-dhango (#2984).
Your answer goes to an agent. Through 0.7.11 some answers acted on the request directly, and some had no receiver when no operator was set. Now these answers use the configured receiving agent or the kernel operator, with the question included. Answers wait when no receiver is available.
rig down now stops only the terminals it started, and names any it left running so you can stop them yourself.rig down is more careful. It stops only a tmux session that carries the seat's own OPENRIG_NODE_ID, on the server and session it read. A terminal OpenRig didn't create, a mismarked one or a replaced one is left running, and rig down exits 2 naming it and how to stop it yourself. A seat joined with rig attach as an external CLI is listed as not stopped, since OpenRig never stops it; that alone doesn't make rig down exit 2.
Codex "Hooks need review" with smol-toml 1.9 no longer needs a manual workaround.
A community harness can declare protocol: pi-rpc, presets can carry launch wrappers, and a seat can switch between Pi and another runtime. The community-maintained nodecode runner is listed in the reference.
Pi and OMP: typed edits submit in order on the affected Node releases (22.21.x to 22.22.0, 24.9.0 to 24.11.1, 25.0.0 to 25.1.0) instead of appending an insertion at the end.
rig down now stops only the terminals it started, and names any it left running so you can stop them yourself.rig down is more careful. It stops only a tmux session that carries the seat's own OPENRIG_NODE_ID, on the server and session it read. A terminal OpenRig didn't create, a mismarked one or a replaced one is left running, and rig down exits 2 naming it and how to stop it yourself. A seat joined with rig attach as an external CLI is listed as not stopped, since OpenRig never stops it; that alone doesn't make rig down exit 2.
openrig-core is 0.1.28. A Codex session kept running across the upgrade can ask for hook approval until it's relaunched.
rig gateway human add. Setup's guidance now asks your agent to offer it, and nothing is registered without your yes.Setup's guidance asks your agent to offer to register you as the team's person. In testing the agent didn't reliably make that offer (0 of 5 runs). Registering works when you ask for it: rig gateway human add. Nothing is registered without your yes, an existing default person is kept, and no person is chosen automatically. The empty-registry hint is right. A team you start from your own assistant fits your own logins, with a review before anything is committed.
Answers nothing else reads, such as a host pairing request or a mode's scheduled end, use the configured receiving agent or the kernel operator. Answers wait when no receiver is available, and nothing decides them meanwhile. To decide one yourself, use its own command: rig host approve or rig host deny for a pairing request, rig seat decide-permission for a runner's request. Closing an ask isn't permission, and words that were kept aren't a default permit.
Pairing answers go to the receiving agent, and you can decide directly in the TUI's pairing settings. A pairing request says who it went to and who reads the answer.
A runner's permission answer goes to the receiving agent, which decides while the runner waits, so a qualified answer such as "only if it's the test file" is read before anything acts. Your answer is kept and an explicit decision works, but the runner's confirmation updates to you aren't written in this release. Don't expect a confirmation in the Feed, a notice of the receiver's state or an immediate wake: the decision reaches the runner at its next read (up to about 21 seconds in testing). To decide one yourself: rig seat decide-permission <askId> --option <id> --reason "…". This is the runner's path only: a pairing request says who reads the answer, as above.
OPENRIG_RELAY_MODE=pull. A community contribution.A relay connector that can't accept incoming connections can pull pending messages for a person instead of being pushed to: set OPENRIG_RELAY_MODE=pull in the relay secrets file. Push stays the default. A person's gateway binding can be unset.
rig terminal open <view> --provider herdr-app. A community contribution.rig terminal open <view> --provider herdr-app opens a Rig view inside the herdr-app desktop window.
Queue content can be erased for closed-work retention and account deletion, and a retried create can't bring erased work back.
Each send records what's known about its delivery in one transport.delivery_evidence event: what was admitted, what reached the input, what the screen check showed, and for which occupant.
rig queue create flags an unresolved scope, rig queue show names a handed-off task's successor, and a custom blocker says what it's waiting for. A community contribution.rig queue create advises on an unresolved scope, slice lists show observed queue counts, a proof policy can require every listed judge, the execution ladder shows each slice's recorded status, rig queue show names a handed-off row's successor, and a custom blocker ID explains the wait.
rig seat status shows when that guidance was written.Every seat's managed guidance points to its skills, help and project context. rig seat status shows when guidance was written relative to the session's recorded moments, and what the seat's refocus hook recorded. A rebuilt seat's briefing lists older recaps as optional history.
rig queue show and rig parked are faster: they no longer re-read every item. A community contribution.Speed: queue listings, rig queue show and rig parked stop re-reading each item; the stuck sweep yields by elapsed time; Markdown section ends are found in one pass.
Delivery and drafts: Claude seats with a custom status line read idle and verify sends; a fresh idle hook doesn't type into a current Claude dialog; Claude's context hint and drawn cursor are read correctly; when the input can't be read, a deliberate override (--dangerously-interact) pastes with a warning and doesn't press Enter, so it isn't submitted; update notices beside the paste hint are recognized.
Codex: when you answer a Codex choice prompt, a message waiting on it is delivered instead of escalated; a working counter is compared only within one pane process; hook trust records are written under smol-toml 1.9, so launches no longer stop at "Hooks need review".
rig daemon start clears a reservation left from an earlier boot, and a Claude agent resumes the right conversation. A community contribution.Restore, start and handover: owed restore input is delivered after a bounded wait on unknown readiness; compaction uses the draft defaults and reports repeated unsent preparation; rig daemon start archives a start reservation left from an earlier boot; the kernel restart guidance describes the existing recovery; an earlier occupant can't replace a Claude seat's resume target; token switches keep the Claude launch; rebuild continuity and Codex launch choices are explained.
Queue and recovery: a polling hosted member (an outside assistant on the team; experimental and off by default) gets one standing notice; a hosted member's ring that times out reads unknown, and durable recovery doesn't repeat that recorded wake (other recovery paths can still ring it again); unroutable items wake as soon as the seat is reachable; the stuck sweep counts a finding once; a handoff doesn't change an escalation's owner; findings route to the seat's own fallback; a park on an outside condition with no wake says so; rig workflow validate is read-only.
Slack and people: a Slack reply that can't be read says the message "may have posted"; threads are recovered when needed; an ask settles after its original message is deleted; eligible asks post when quiet hours can't be read, marked when OpenRig can't fully check them; a held ask whose person has no connector closes as no-connector; update delivery is described without assuming Slack.
rig context recap-write writes where every reader looks, the installer tells you how to fix PATH, and daemon credentials stay out of a tmux server's environment from its next start. A community contribution.CLI and config: rig context recap-write writes where every reader looks; config accepts two keys the daemon already resolves; rig whoami says the team is still being built during rig up; rigs with @ in the name find attention; a seat waiting on a person names its question; the installer reports PATH gaps with shell hints; daemon credentials stay out of a tmux server's environment from its next start (a server already running keeps its old environment until it restarts).
Terminals: a Herdr grid whose blank filler cells have no label is confirmed, instead of reading every seat unconfirmed.
rig down leaves alone any terminal it can't show belongs to the agent.rig down leaves terminals it can't show belong to the seat; see Before you upgrade.
rig daemon start now archives a start reservation left over from a reboot, so you no longer clear it by hand. A community contribution.A daemon start reservation left after a reboot is archived by rig daemon start; don't clear it by hand.
rig down: a terminal it leaves running stays running until you stop it. It doesn't keep a stored host or boot identity, doesn't guard against another program running as you, and a daemon restart between losing the tmux server and the stop wasn't exercised. (#2962)agents.operator_session) that runs in an explicit or auto permission mode gets no OpenRig allowance for rig host approve or rig host deny, so Claude's review can refuse them. A kernel receiver in auto mode is included. The default kernel with the default policy passed at its tested scope.AGENTS.md restore; Codex account-home, upgrade and continuity checks; the OpenShell worker's second report; Pi model switches on real providers; ACP in a desktop app; Slack directory paging; some Slack setup cases (channel and thread precedence, a stopped receiver's delayed pickup, an inherited inbound default, a plain yes with no receiver); native checks of #2786, #2799, #2807 and #2813, with #2947 partly; a button-and-words answer, a non-pairing ask with an explicit owner, Steering's Approve page, and these answer and setup checks on a default model (the tested runs used smaller models); --wait-for-kernel (#2888), the installed effect of #2800, and #2720 beyond one platform.Thank you to everyone whose work is in this release: @billbenetti, @oodadoudou, @DeryFerd, @Aummadour, @seanmcn, @w4rc0n, @ziadeh, @vanducng, @smota, @sn1thy, @dajiaohuang and @Vaishnavi220506; to @KM199, co-author of #2947 and #2952, which finish their #2513 and #2515.
Thanks also for the reports behind these fixes: @nassersala (#2955), @djuntgen (#2804), @todd-dhango (#2851), @Kayhusk (#2796), @pr0xc3nt4ur1 (#2849 and #2850), @amsayfi (#2776), @oodadoudou (#2847) and @CaptObvious (#2933 and #2942).
Late verification update: a real Codex lifecycle journey passed its tested launch, same-conversation snapshot restore, existing-rig resume, Core restart with native sessions retained, and fresh-handover task cases. The packaged notes still list that journey as not run because this result arrived after their seal. This does not establish all models or all bidirectional team workflows.
Known limitation: a Codex snapshot restore or rig up --existing can write OpenRig's managed guidance blocks into a tracked AGENTS.md, leaving the worktree modified. Check git status after either operation. Snapshot restore was already disclosed in 0.7.11; the existing-rig trigger is newly observed, and its regression status is not established. A repair is selected for 0.7.13.
The Codex account-home smoke is partial: launch passed, while the account switch was not run after its dry-run requested a named saved profile. Codex ring-2 upgrade coverage and the Codex half of the person-setup checks remain not run at this release decision. The other limitations and coverage omissions in the release notes still apply.
The openrig plugin, shipped beside openrig-core, lets a person hand work to their local team from their own Claude Code or Codex, follow it, answer the team's questions and read the result. The plugin's README has the steps. VS Code, Copilot CLI and Cursor install it with no seat. Answers are recorded exactly as typed.
rig acp serve makes it an ACP agent, so a prompt becomes a task and the team's answer comes back as the reply.rig acp serve --to <member@rig> runs OpenRig as an agent in ACP apps such as Zed and T3 Code: a prompt becomes a queue task, each task the work reaches shows as a step, the closing resolution is the reply, and the team's questions reach the app while a turn is open.
rig config set transport.draft_delivery refuse.Under the default deliver, an ordinary send or work notice to a seat whose prompt holds a person's typed draft pastes the message at the cursor and submits the combined input, and the result says whether submission was confirmed. Menus, permission prompts, the typing guard, managed compaction and --wait-for-idle still refuse or wait.
/compact at the next idle moment instead of asking again. A community contribution.A prepared managed compaction is held as prepared and sends /compact at the next idle moment instead of asking the seat to prepare again. rig compact --state shows it and --cancel disarms it. Managed compaction also sends /compact under Claude Code's bare update banner.
On Linux, a daemon run as a systemd user service starts the seats' tmux server in its own user scope, so stopping or restarting the daemon's unit no longer ends the seats. A daemon started from a login session starts the server as before.
Run npm install -g @openrig/cli@0.7.11, then rig daemon stop && rig daemon start. Back up your database first. Five migrations run at the daemon's first start (150 to 154). With the daemon stopped, copy the file $OPENRIG_DB names. Going back to 0.7.10 after these migrations isn't established.
The openrig-core plugin moves from 0.1.26 to 0.1.27. Its activity relay and refocus hooks and three skills changed. The CHANGELOG says replaced files are kept under $OPENRIG_HOME/backups/managed-refresh/<id>/ for reapplying local edits; this release's checks didn't test that route.
rig config set transport.draft_delivery refuse.Sends now join a person's draft (transport.draft_delivery = deliver). To keep the old refusal, run rig config set transport.draft_delivery refuse.
PI_CODING_AGENT_DIR in its env: is used as one and needs an auth.json there. A Pi seat that already has its own login can stop at launch, so check those first. A community contribution.A Pi seat that declares an absolute PI_CODING_AGENT_DIR in its env: now uses it as its account home and needs a readable auth.json there, or the launch stops. A Pi seat that already has its own Pi login also stops at launch when its spec declares an absolute PI_CODING_AGENT_DIR, even if that folder has an auth.json; the seat's login is kept, and the error names a missing account file rather than the existing login. This is shown by source and function-level tests, not a native run.
Linux, with the daemon run as a systemd user service: the seats move into their own scope only when their tmux server next starts fresh. Until then, stopping or restarting the daemon's unit still takes the seats down.
A draft only the input box shows is caught by one read before pasting.
Claude seats showing a statusLine or the background task panel read as a normal composer under the draft-aware guard.
A full held-message quota no longer drops a sender's later messages, and seats stopped while they owe work don't get a burst of undeliverable notices at each daemon start.
on_restore: follow_source. A community contribution.on_restore: follow_source on a skill refreshes it from its source when a seat is restored; keep_saved stays the default.
/compact instead of a refusal.The final /compact allows for the preparation's own fresh activity (23 seconds, against 9), so a seat that's already idle gets /compact instead of a refusal.
A restore stopped at a prompt, such as Claude's resume selection, clears itself once the seat runs the conversation it was restoring.
A work notice refused at a prompt no longer goes in right after "Compacted" while the compaction still owes its restore request.
A Codex helper inside a Claude seat no longer becomes the seat's resume token; a Claude rebuild handover checks its priming packet; restoring activity hooks finds the agent's own source.
@openrig/acp launcher runs the matching CLI's rig acp serve through npx.The @openrig/acp launcher runs the matching @openrig/cli's rig acp serve for the ACP registry's npx launch, released at the same version.
Assistants connected through a grant service can list and answer the team's questions as a registered person, and --task-destination names the receiver for new tasks.
The relay gateway connector sends questions, updates and notices to an operator-owned bridge over loopback HTTP and takes replies back. The bridge is yours.
The web UI's For You feed answers a local ask's questions on its card.
--host, and whoever held a task hears when another machine takes it over. A community contribution.rig queue show, list, transitions, update and block take --host <id>; cross-host handoffs that override a claim notify the claimer.
rig queue artifact knows what to diff.An openrig.code-result/v1 manifest gives rig queue artifact the commit pair to diff.
rig workflow unregister removes old versions, and trace --full shows the whole trail. A community contribution.Workflows keep their own spec and root objective, rig workflow unregister removes versions, and trace --full returns the whole trail.
One backlog notice replaces a finding per task for a seat that reads without claiming; claims read working while a Claude owner has a live background shell or a managed restore is owed.
workspace.root. A community contribution.A helper harness inside a seat no longer reports its turns as the seat's; refocus traces projects outside workspace.root.
rig provider switch, and Pi seats with their own account home start and resume again; the switch is new, so check the seat afterwards. A community contribution.rig provider switch --seat <pi seat> --to-account <name> --account-home <dir> --reason "<why>" moves a Pi seat to another account through a confirmed successor (--dry-run shows the plan); in our native run the successor started but the switch wasn't completed, so a completed switch is untested. A Pi seat whose spec declares its account home starts and resumes again.
--settings JSON that contains spaces verify again, and launcher scripts that exec Claude's symlink are recognised. A community contribution.A Claude seat launched with --settings JSON written with spaces can be verified again, and a Claude launcher script that execs the native installer's symlink becomes ready, with --no-chrome recognised; in our test, 0.7.10 also became ready under the same chain, so this isn't a demonstrated fix.
launch_wrapper works for Claude Code, Pi and Codex seats, and a wrapped seat never starts without its wrapper. A community contribution.launch_wrapper works for local Claude Code, Pi and Codex seats, and a wrapped seat never starts unwrapped.
Claude seats that load openrig-core themselves stop loading its skills twice.
rig attach --self outside tmux, rig host pair --timeout, clearer rig config set warnings, and steadier daemon start and stop. A community contribution.Also: rig attach --self outside tmux, rig host pair --timeout, rig config set naming a masking variable, daemon start and stop treating a blocked probe as unknown.
Answers to a pairing request now reach the target's kernel operator. When the target has an operator that can receive them, typed words and Approve or Deny choices in the TUI, Steering or Slack go to that agent as replies, and receiving one neither grants nor denies access. The operator acts on clear intent with rig host approve <code> or rig host deny <code> and asks back when the answer is unclear. Both commands also work from a plain terminal on the target, using its operator bearer (OPENRIG_AUTH_BEARER_TOKEN, or --bearer <token>). An Approve click on a request the operator owns waits for the operator to act; a stopped operator keeps the request, and the answer waits in its queue.
Checked on this package: real Slack with a real operator agent in one run (a button or a typed yes approved, a no denied, and a qualified reply got a clarifying question before the refusal denied it); the real TUI without a model (answer credentials, typed answers asked again with no operator, plain-terminal approve and deny, and expiry); and 29 of 29 contained cases with synthetic credentials. Not run: an agent reading answers given in the TUI, Slack thread replies beyond the tested path, and Steering in a browser.
With no operator, one that can't receive answers, or one OpenRig can't determine, the request says so: Approve and Deny decide directly, and a typed reply gets another ask instead of a denial (the no-operator case was checked; the other two come from the source).
TUI answers use the daemon's operator bearer when one is configured; a missing or wrong one leaves the answer unrecorded and says what to set. Ordinary terminal calls keep their own credential.
Denying or dropping a host pairing request in Steering now gives the requesting machine no access. The problem, that action returning the daemon access token, was found in the 0.7.9 and 0.7.10 source and reproduced with synthetic credentials on an earlier 0.7.11 build.
Herdr follow views work on macOS too (protocol 22).
Writable tiles reflow with the screen: they attach as in 0.7.10, so the screen in use sizes the seat and text reflows on resize, and once attached tmux handles typing and paste directly. Read-only tiles keep showing the seat without attaching, typing into it or resizing it. Fixed web geometry is opt-in (terminal.fixed_geometry, off by default); that browser option wasn't exercised in our checks. Checked in one run in real Herdr 0.9.5 on tmux 3.4 (reflow while viewing, typing and paste once attached, and a tmux window fixed by hand at 80x24), with typing and paste also checked on tmux 3.4, 3.3a and 3.6; phone-size screens and tmux 3.3a weren't run in Herdr.
Closing the herdr-core terminal that started OpenRig no longer ends the daemon or its seats. After merging, @w4rc0n ran a real close with herdr-app 0.1.13 in an isolated Linux sandbox: without the change the daemon, tmux server and seat ended, and with it they kept running. That sandbox had no systemd --user, so herdr-core's service-unit path wasn't run, and a tmux server started inside the terminal before the daemon still ends with it.
rig doctor says why tmux control mode is off. A community contribution.Terminal snapshots keep colours, a reattach during a broker teardown works, and rig doctor says why tmux control mode is off.
The stuck sweep batches its reads: on a copy of a busy 340 MB database, a pass went from about 11 s to about 5 s and the longest stall from about 10 s to 0.57 s.
view.changed isn't stored any more, and retention.events_days can prune history-only events, which can't be undone. A community contribution.view.changed events are no longer stored, and opt-in retention.events_days prunes eight history-only event types.
daemon.seat_poll_interval_ms lets you trade activity freshness for daemon CPU. A community contribution.The web UI loads the newest 100 events; daemon.seat_poll_interval_ms trades activity freshness for daemon CPU.
Docs and tests. Daemon HTTP clients and bearer tokens, the 0.7.10 capability map, a neutral TUI demo snapshot, a web UI editor timer, and test and CI fixes.
rig config set transport.draft_delivery refuse brings the refusal back.Sends and work notices that were refused with prompt_draft now go through by default, joined to the person's draft. rig config set transport.draft_delivery refuse restores the refusal from the next send, and under refuse a draft only the input box shows now refuses too.
PI_CODING_AGENT_DIR uses it as its account home, and won't launch without an auth.json there. A community contribution.A Pi seat with an absolute PI_CODING_AGENT_DIR uses it as its account home; no auth.json, no launch.
/api/events no longer sends view.changed. A community contribution./api/events no longer emits view.changed.
The stuck sweep files one backlog notice where it filed one per task.
rig compact over it reports already_in_progress. A community contribution.A prepared managed compaction stays waiting, and a manual rig compact over it reads already_in_progress.
HC_SCOPE, HC_TERMINAL_ID or HC_TERMINAL_SOCKET. A community contribution.The daemon and seats no longer inherit HC_SCOPE, HC_TERMINAL_ID or HC_TERMINAL_SOCKET.
Read-only terminal tiles show the seat without attaching; writable tiles attach as in 0.7.10, and detaching ends that viewer while the agent keeps running.
rig host approve or rig host deny.On a target with a kernel operator that can receive answers, an answer to a host pairing request no longer decides it: the operator decides with rig host approve <code> or rig host deny <code>.
✔ Update installed banner can paste the message and not press Enter, leaving it typed but unsent. Look at the input box before sending again. A fix is on main (#2786) and isn't in 0.7.11. Managed compaction under the same banner sent /compact 3 of 3 times in our test.rig compact <seat> --skip-map is the supported manual recovery route, subject to its normal idle and delivery checks; this route was not exercised by the 0.7.11 release checks./compact. Run rig compact <seat> --cancel, then send a restore request naming the newest restore map, marker and packet. A work notice refused at a prompt now waits for that step, but notices at daemon start and on a new handoff don't check for it yet. (#2575, #2666)delete-owned) now works for a custom image, and recreate and an exact resume passed in our native run. A restored guest takes in its startup content again, so strict recall after a restore isn't met. In our run, finishing the guest's task was assisted and time-boxed, so whether it finishes unassisted is unknown. Never give the guest the daemon's main API port (normally 7433), and after a same-generation reconciled launch, run rig environment connect yourself. (#2755)rig down can start a duplicate claude --resume for a moment. It's pre-existing and the same on 0.7.10, and we saw no conversation write from it.refuse, an unreadable input box lets the send go ahead, so an undetected draft goes with the message; under deliver, a result may not mention a draft it joined. (#2645, #2653)draft_input_unknown still isn't retried. (#2605)[file reply], keeps no words there and doesn't deny (from the source; not run). (#2900)rig host approve or rig host deny sends its outcome directly. A request that expires unanswered gets no further update. This comes from the source; no restart case was run. (#2900)env -u ordering is tested in macOS CI; for GNU and BusyBox it rests on their documented behaviour, because the Linux CI job doesn't exercise that branch. (#2782)rig terminal open; for a --follow view, use --no-follow first, then open a fresh view. The agent and its conversation aren't touched. In our run, kept 0.7.8 and 0.7.9 views kept their process, session and history through the upgrade and the refresh, but they already reflowed before it, so this isn't shown to repair a frozen viewer, and the follow-view upgrade didn't confirm, so it isn't qualified. (#2868)keep_saved skills: a skill whose source changes between rig down and the restore comes back missing, named in the restore's warnings; an edited seat copy is kept. (#2626)rig exec: checked with Claude Code and Codex through their ACP adapters. A cancel is acknowledged, not settled, and says nothing about further spend. Cancel with rig exec cancel, not rig acp cancel.openrig plugin hasn't run in Claude Desktop, the Codex app or Copilot CLI. A seatless app can't claim, update, hand off or answer questions, and everyone using the app on a machine acts as its seat. (#2647, #2657, #2683)launch_wrapper hasn't run with zsh, on macOS arm64 or with a real installed Claude, and an alias or shell function named claude stops the launch. (#2746)rig seat restart --keep-conversation refuses a Codex seat launched with developer instructions (settings_drift), and one started by a 0.6.x tmux-send script (#2200).hooks.json, OpenRig can leave its activity hooks and trust records unchanged. The range (^1.8.0) still permits 1.9.x; the repository lockfile uses 1.8.0. Not fixed in 0.7.11; a fix is on main (#2804) and isn't in this release. (#2720)AGENTS.md. Check git status before committing.unknown behind ProxyJump or ProxyCommand; advisories don't show on a native resume; seats handed over before 0.7.9 refuse stop, restart and clean; after a project-local install, plain rig isn't found in the TUI's own shell; with Claude Code 2.1.319, rig up can wrongly say "Startup prompt still staged … press Enter"; two npm installs into one folder at once once left a partial install, cause unknown.docs/reference/seat-delivery.md says drafts are protected without naming the setting. Under the default, transport.draft_delivery: deliver, an ordinary send that meets a person's draft is joined to it and submitted, unless the seat's typing guard holds input (draft-aware, openrig-urgent, urgent-only or hold).--binding on rig gateway human add is optional, though the CLI reference marks it required.--delivery-class (for example --delivery-class B).Thank you to everyone whose work is in this release: @billbenetti, @seanmcn, @smota, @Aummadour, @dajiaohuang, @w4rc0n, @Mohsintl, @twotnguyen, @ziadeh and @Nikhi00718; to @w4rc0n for running the real terminal close behind #2737.
Thanks to @billbenetti for the reports behind #2752, #2660, #2738, #2749, #2666 and #2665; to @pr0xc3nt4ur1 for the report behind #2740 and #2743; to @oneandonlydean for the process chain behind #2751; to @niklasbaldi for the reports and causes behind #2654 and #2671; to @vadym-abstract-factory for the report behind #2717; to @w4rc0n for the fix and root cause in #2821; to @ziadeh for the reports behind #2622 and #2486; and to @mlc-factory-bot for the reports behind #2605 and #2607.
rig exec, and a step held after compaction now goes in instead of waiting forever.3 added13 changed11 fixedrig exec run: it runs one turn and reports back, and you decide whether to accept the work.rig exec run <row> --harness claude|codex runs one turn through the maintained ACP adapter, in a connector that outlives the command. Use --plan, --wait, status and cancel. One run per row, and --again starts another turn. The outcome is the agent's report. Accepting the work stays with you.
rig compact <seat> --state says when that happened.After two idle polls that can't read the input box, the held step goes in, and while a later step is still owed, rig compact <seat> --state says the last one went in over a box it couldn't read.
Delivery leaves every tmux mode, and a refusal says why.
The rigs skill tries a user-writable npm prefix or a version manager's Node before a project-local install.
uses key gets one. A community contribution.Spec advisories show as launch warnings and never stop a launch, and an unknown uses key gets one.
Run npm install -g @openrig/cli@0.7.10, then rig daemon stop && rig daemon start. There are no migrations.
The openrig-core plugin moves from 0.1.25 to 0.1.26. It adds a permission-request hook and changes the refocus hook. Replaced files are kept under $OPENRIG_HOME/backups/managed-refresh/<id>/, so reapply any local edits from there.
rig agent validate <path>: a single-value uses.<category> now selects that resource, so a member fails to resolve if its pool doesn't hold it.Run rig agent validate <path>: a single-value uses.<category> now selects that resource, so a member fails to resolve if the pool doesn't hold it.
guest.py. Its cleanup still fails against OpenShell 0.1.6.OpenShell users: rebuild the guest image with this release's guest.py (see daemon/assets/openshell/README.md).
A stale record naming a reused tmux pane no longer stops a send from leaving the mode.
Writable views size the pane they show.
Herdr views kept from 0.7.5 through 0.7.9 keep working after upgrading (Linux Herdr 0.9.5 / protocol 22): when only OpenRig's tile script changed, a viewer stays confirmed and a follow view goes on adding and closing tiles. A kept viewer that's already running keeps its earlier sizing for as long as it runs, its own re-attaches included. With a live Herdr after upgrading from 0.7.9, a kept viewer and a follow view's join worked. Closing a tile when a seat leaves wasn't confirmed. It follows @natefaerber's approach in #2467.
rig seat status shows a Claude seat's latest permission request, with suggested rules labelled. Nothing is applied or approved, and suggested rules can contain sensitive values. A community contribution.rig seat status shows a Claude seat's latest permission request, with suggested rules labelled. Nothing is applied or approved.
rig broadcast says UNCONFIRMED instead of FAILED for an uncertain recipient, refusals for seats handed over before 0.7.9 say why, rig roster reads conflicting activity as unknown, and set-runtime finds pod recaps. A community contribution.Routing and status: a send to <seat>@<rig>@<this host's ID> stays local; rig broadcast says UNCONFIRMED, not FAILED, for an uncertain recipient; refusals for seats handed over before 0.7.9 say why; rig roster reads conflicting activity as unknown; set-runtime finds pod recaps.
Refocus names a trace gap instead of failing.
The stuck sweep stops filing informational updates to people as unclaimed work.
rig slack status counts cross-host relays still in flight. A community contribution.rig slack status counts cross-host relays in flight.
rig works in managed Codex seats however OpenRig was installed, with a Codex that supports allow_login_shell, because tool commands now run in a non-login shell, so login-only shell setup may be missing.Plain rig works in managed Codex seats however OpenRig was installed, with a Codex that supports allow_login_shell: tool commands now run in a non-login shell, and rig config set launch.codex_login_shell true defers to Codex's own setting.
rig doctor gives up on a silent listener after 2 seconds. A community contribution.Errors are clearer: a refused terminal token says "Terminal auth failed"; the TUI shows the daemon's launch refusal; rig doctor gives up on a silent listener after 2 seconds.
The execution view reads the review registry once per request. Queue listings build one recovery index instead of a scan per listed task. The event drain reads at most 500 rows at a time. On a test fixture, full listings were about 63% faster with the same output. The listing and drain changes are unmeasured on a live install.
The guest passes the exec's CA settings to Claude, so API calls stop failing with SELF_SIGNED_CERT_IN_CHAIN. TLS checks stay on. With OpenShell 0.1.6 the guest now authenticates, does useful work, makes a verified export and completes its task.
environment_cleanup_unconfirmed.Owned-environment deletes read Podman's inventory, and an unconfirmed cleanup stops as environment_cleanup_unconfirmed. The maintained cleanup still fails against OpenShell 0.1.6 (see Known issues).
Docs and tests. Codex named profiles, openrig-core skill loading, what the runner contract has run, telemetry's guarantee beside the separate, untested openrig-telemetry exporter, and a real-tmux test that waits for the prompt.
A held post-compaction step goes in over an unreadable box, so an unrecognised draft may receive the restore text.
Sends leave any tmux mode a person has open.
rig up can print advisory warnings.rig up can print advisory warnings.
login: true gets Codex's refusal.Managed Codex seats run tool commands in a non-login shell, so login- or rc-only setup, shell functions and aliases may be missing, and a model request for login: true gets Codex's refusal.
delete-owned) fails against OpenShell 0.1.6, which doesn't mount a workspace volume, so recreate and restore can't run. Never give the guest the daemon's main API port (normally 7433). After a same-generation reconciled launch, run rig environment connect yourself.rig exec: it was checked with Claude Code and Codex through their ACP adapters. A cancel is acknowledged, not settled, and says nothing about further spend. Cancel with rig exec cancel, not rig acp cancel./compact. Run rig compact <seat> --cancel, then send a restore request naming the newest restore map, marker and packet. (#2575)✔ Update installed · Restart to update banner, and a message held that way isn't replayed. Managed compaction can also stop before sending /compact (target_activity_unknown) on a Claude seat showing that banner; in our test, a freshly launched seat once let it through, which isn't a tested fix. (#2605)execs the native installer's symlink can still lead to a readiness timeout even though Claude is visible. The launch check also does not recognize --no-chrome. This pre-existing issue is not fixed in 0.7.10. (#2341)rig seat restart --keep-conversation refuses a Codex seat that was launched with developer instructions (settings_drift on developer_instructions) and stops nothing.hooks.json in the Codex home, OpenRig can leave its activity hooks and trust records unchanged and log "its trust records couldn't be written exactly". The reporter saw Codex stop at "Hooks need review". The dependency range in this release permits the affected version; our repository lockfile uses 1.8.0. This is a pre-existing issue, not fixed in 0.7.10. (#2720)AGENTS.md, even on Codex 0.164 or later, leaving the working tree dirty. Check git status before committing.unknown behind ProxyJump or ProxyCommand, which doesn't mean the connection is broken.rig isn't found in the TUI's own shell (managed Codex seats are fine). Use the full path.rig up can wrongly say "Startup prompt still staged … press Enter". Pressing Enter on the empty box adds nothing.Thank you to everyone whose work is in this release: @dajiaohuang, @Aummadour, @billbenetti, @smota, @Nikhila-Dhevi-R and @Vaishnavi220506; to @AlephCasara, whose work on a fork pointed us to the unpaged event notification drain (#2610), and to @natefaerber, whose approach in #2467 the Herdr view fix follows; and, for reports carried into this release, @billbenetti, @KM199, @ziadeh, @dajiaohuang, @jajabong, @george661 and @aanestesia.
Thanks also to @smota for the OpenTelemetry discussion behind #2589.
Before release we tested the exact package: the private and stock content checks and the release Tests; a real upgrade from 0.7.9 on our own servers; the speed and memory baseline (no worse, slightly faster); the OpenShell worker's sign-in, useful work, verified export and finished task (its cleanup still fails, and recreate and restore weren't run); Slack on a real workspace; a Claude first user from the website to a finished result; Codex first-user setup (plain rig in every Codex seat); and 98 selected broader regression cells (57 passed, 8 had other results, 14 were incomplete and 19 weren't run), with no product regression found. A Codex first user didn't reach a finished result in our test: both attempts stopped on errors in our test setup, so that path's end result is untested.
rig stream emit opens a queue task for the kernel's queue worker. To keep 0.7.8's behaviour, run rig config set stream.intake off after installing. A community contribution.Stream intake is on by default. Each rig stream emit now opens an intake row for the kernel queue worker (queue.worker), or adds a note to the one still unclaimed. The row wakes the queue worker like any queue row, and earlier stream history isn't replayed. To keep 0.7.8's behaviour, with no intake rows, run this after installing: rig config set stream.intake off.
A person's typed words reach the agent, and the ask says what they do. A reply of just yes or no counts as the agent's allow or reject. "Any other words choose nothing": OpenRig's default permits the request when the agent offered an allow option, and the words come back whole in the read result as said. Section 7 of docs/reference/harness-runner-contract.md lets a runner queue them for the agent "as an ordinary prompt" after the current turn: "They're never the answer."
rig stream list --tag runner-permission-permit.Each permit OpenRig grants by default is recorded, once, as a stream item from the seat: when no one can be asked, and when an ask is closed without a choice (a row passed to an agent seat included). The record names the tool call, why no person answered, the option the agent got with its kind, and the request or ask. Read them with rig stream list --tag runner-permission-permit. A record has no destination, and one that can't be written doesn't change the permit. Stream intake reads past it, so it opens no intake row and wakes no one.
A working beat keeps a long, quiet turn reading working (section 4); a runner that sends none behaves as before. The Pi runner and the dsh example post it, and the conformance check covers it and no longer falsely fails a correct runner on a busy machine.
Delivery: fewer deliveries held or misdirected on Claude seats (prompt suggestions, wrapped lines, a wake's own paste, the Theme picker, quoted permission questions); held messages are announced when a seat starts or changes occupant; undelivered reminders retry; a stalled tmux client is ended after 30 seconds.
Restart the daemon after installing: run npm install -g @openrig/cli@0.7.9, then rig daemon stop && rig daemon start. Four migrations (146 to 149) run on the first start.
rig stream emit opens a queue task for the kernel's queue worker. To keep 0.7.8's behaviour, run rig config set stream.intake off after installing. A community contribution.Stream intake is on by default. Each rig stream emit now opens an intake row for the kernel queue worker (queue.worker), or adds a note to the one still unclaimed. The row wakes the queue worker like any queue row, and earlier stream history isn't replayed. To keep 0.7.8's behaviour, with no intake rows, run this after installing: rig config set stream.intake off.
The bundled openrig-core plugin advances from 0.1.24 to 0.1.25: its seat-continuity-and-handover skill changed. Managed refresh preserves replaced files under $OPENRIG_HOME/backups/managed-refresh/<id>/; reapply local edits you still need from that copy.
If you run an OpenShell environment, rebuild its guest image with this release's helper (daemon/assets/openshell/README.md). An older helper rejects the new startup step, and the error names the rebuild.
Experimental and unsupported. Off by default. Use at your own risk. This path needs a grant service that you supply and run; ordinary use of OpenRig doesn't need one. With rig mcp serve --http --grant-service <url>, a connected assistant can now claim, update and park the rows addressed to its own seat, and see its team with rig_team, within the scope its existing tools have. Serving MCP over HTTP behind your own tunnel "is like opening a port in your firewall." See docs/reference/hosted-members.md.
New daemon reads, two of them by @dajiaohuang: people and the default person, a person's open updates without Slack, a rig's messages in one request, and Slack channels and people without the client holding the bot token. An opt-in decisionRevision lets a client settle exactly the decision its person reviewed. docs/reference/daemon-http-clients.md covers them. The people, updates and messages reads pass their route tests, but no real client used them in our release testing.
specs.builtins_visible: false hides built-in specs, a Claude member's plugin_channels pass to --channels (not yet run with a real channel plugin), and rig context add <url> installs into a new nested name. A community contribution.Options and views: specs.builtins_visible: false hides built-in specs; plugin_channels passes a Claude member's plugin channels to --channels (by @Aschenbath, carrying on @todie's fork work), warns beside claude_mcp: none, and hasn't been run with a real channel plugin; rig context add <url> installs into a new nested name; TUI and web UI display fixes.
Status and receipts: idle Claude and busy Codex seats read correctly in more layouts; stale window samples expire, so a seat whose pane can't be read reads unknown rather than its last verdict; uncertain sends read unconfirmed or indeterminate instead of delivered, lost or failed.
CLAUDE.md and AGENTS.md files stop growing, and several Claude rebuild and resume failures are repaired.Guidance and handover: an edited culture or startup file reaches the next start, while your own edit to a managed block is kept; shared CLAUDE.md and AGENTS.md files stop growing; handover successors get their guidance and a list of their startup files; several Claude rebuild and resume failures are repaired.
rig, rig doctor no longer hangs on a silent port, and upgrade backups work without the sqlite3 CLI.Install and hosts: the upgrade skill backs up without the sqlite3 CLI when Node has node:sqlite; preflight warns when a Codex seat's login shell can't find rig; rig host ls reads your ssh config with ssh -G, which runs any Match exec command in it; rig doctor no longer hangs on a port held by a listener that never replies; a scoped launch on tmux 3.4 keeps its default command.
The OpenShell worker still doesn't work in 0.7.9. Don't rely on this path. 0.7.9 repairs the layer that stopped it in 0.7.8: in our release testing the guest's agent now starts and reaches its first input. Startup reaches the guest through a hash-checked first-input file, and nothing is typed through the host viewer; the case where the viewer exits wasn't run in our release testing. Readiness checks the guest's own execution, and a failed owned delete reports the stage that failed. Our release testing on 0.7.9 then found two more problems: The worker can't reach its model. The guest's Claude starts, but every API call fails with SELF_SIGNED_CERT_IN_CHAIN. The guest launcher builds a clean environment and leaves out the CA settings the OpenShell exec supplies (NODE_EXTRA_CA_CERTS, SSL_CERT_FILE, REQUESTS_CA_BUNDLE and CURL_CA_BUNDLE). Deleting an owned environment doesn't run. Its inventory step misreads a "Not Found" reply, so the delete is never sent and the owned resources stay. In our test they were removed with the vendor's own cleanup. Attached resources were kept, as intended, and the worker stopped cleanly. With the four CA settings passed through by hand, a worker did useful work; a verified export, an exact restore and a finished task weren't shown. Fixes for both problems are on main (#2602, #2603). They aren't in 0.7.9 and haven't been run end to end.
rig config set stream.intake off keeps 0.7.8's behaviour.Stream intake is on by default (see Before you upgrade).
AGENTS.md.Fresh Pi seats, and fresh Codex seats on 0.164.0 or later, get OpenRig's guidance outside the repository AGENTS.md: Pi in its own agent folder (unless that file or folder is a link), Codex as launch developer instructions. Codex still uses the repository file for a resume, fork, restore or handover successor, an older or unknown version, or a config layer that sets developer_instructions or can't be read. If only your shell startup sets CODEX_HOME, OpenRig reads ~/.codex, so a developer_instructions in your home is replaced for that session; declare the same absolute path as CODEX_HOME in the member's or rig's env: to have it checked.
rig down removes OpenRig's own additions from a working folder once no live seat uses them.rig down removes OpenRig's own additions from a working folder once no live seat uses them.
After upgrading, the Slack connector edits asks that ended earlier to show they were handled, at Slack's edit rate: about an hour for 450 asks. rig slack status shows pending edits.
qitem_already_terminal and records nothing.A later answer to a settled ask gets 409 qitem_already_terminal and records nothing.
rig send --verify reports unconfirmed and exits 1 when it can't see what its Enter did. A community contribution.rig send --verify reports unconfirmed and exits 1, not delivered, when it can't see what its Enter did.
rig export with no -o writes rig-export.yaml beside an existing rig.yaml instead of replacing it. A community contribution.rig export with no -o writes rig-export.yaml beside an existing rig.yaml instead of replacing it.
rig workspace validate exits 1 for malformed optional frontmatter. A community contribution.rig workspace validate exits 1 for malformed optional frontmatter.
rig env lists the matching rig IDs and stops, and rig doctor --spec reports unknown, instead of using the first match. A community contribution.An active rig name that matches more than one rig makes rig env down, status and logs list the matching rig IDs and stop, and rig doctor --spec report conformance as unknown, instead of using the first match.
<workspace.root>/kernel; existing kernels stay where they are.New default kernels start in <workspace.root>/kernel. Existing kernels stay where they are; docs/reference/agent-startup-guide.md describes the optional move.
rig environment connect yourself; non-text startup actions aren't supported on environment seats.truncated.unknown behind a ProxyJump or ProxyCommand; that isn't proof the connection is broken. (#2488)CLAUDE.md or AGENTS.md that already grew isn't trimmed.✔ Update installed · Restart to update on its own line. 0.7.9 still doesn't recognize that bare banner, though it recognizes /rc and the full versioned banner. A message the draft-aware delivery path holds isn't replayed automatically by that path. (#2605)rig compact <seat> --state and rig seat status show the restore as owed, and from the second failed attempt the seat's owner gets a queue item naming it. To restore by hand, send the seat a restore request naming its newest restore map, its restore marker and its packet, then settle the owed attempt with rig compact <seat> --cancel; a poll can still race that step. Reported on 0.7.8, and 0.7.9 behaves the same way. (#2575)rig seat stop, restart and clean refuse them as before. (#2440)rig broadcast: an uncertain recipient's outbox receipt now reads indeterminate, but the broadcast's own output for that recipient still says FAILED. Check that seat with rig capture before resending. (#2507)rig: command not found when rig isn't on the login shell's PATH: with Node outside it, or when npm can't install globally and OpenRig is installed inside the project instead. Preflight now warns at launch. In our testing the agents recovered by calling rig by its full path; the getting-started guide describes a ~/.profile fix, which we didn't run.deepseek/deepseek-v4-flash through dsh's pi-ai OpenRouter chat-completions profile, against 0.7.8 Core: two useful queue tasks, queued and multiline input, a quiet tool interval of about 26 seconds that still read working, allow-once and reject-once with their file effects, cancel, a return to the same conversation with recall, and shutdown. The permission answers came from an automated test operator, and the run's preset added resume: true, which the shipped preset doesn't declare. dsh's own DeepSeek Messages route through OpenRouter failed with HTTP 400, and the cause isn't established. Not run: a missing-conversation refusal, cross-host restore, OpenShell, and other adapters, models or providers. 0.7.8's release tests had already run its dsh example end to end, so the 0.7.8 CHANGELOG's "No harness has run end to end under the contract yet" was out of date.env: names, the conformance check's false failures, a person's typed words, the README's operator address, a rebuild behind a symlink, the rebuild receipt, a click on a re-parked ask, and recording a permit made when no one can be asked: a person's answer is recorded on the ask, and a default permit as the stream item above. The other 0.7.8, 0.7.7 and 0.7.6 known limits still apply.Thank you to everyone whose work is in this release: @porterchild, @Mohsintl, @Nikhi00718, @oodadoudou, @AdityaxDeore, @w4rc0n, @genius8267, @Aschenbath, @DeryFerd, @mohitdixit02, @Vaishnavi220506, @Nikhila-Dhevi-R, @natefaerber, @dajiaohuang, @Aummadour, @KM199 and @Totopo27; to @todie and @seanmcn, whose fork work is carried in #2439 and #2400; and, for reports carried into this release, @ziadeh, @billbenetti, @eisenbergjerry, @ch-davidriott, @puniserv, @oneandonlydean, @sabnanikl-dev, @gifflen, @pvl-agent, @EZotoff, @adam-nectar, @bpamiri, @tgrundtvig, @aanestesia, @AdamWhitehurst and @eisenrigbot.
Thanks also to @z4cc, @hitianqi, @jimallen, @m3ac-AllbrittenJ and @lkraider for reports and field feedback.
rig seat set-permissions-to chooses someone else, a request is permitted when no one can be asked, and refusing is opt-in with --deny.A runner seat's permission requests now reach a person, and the default is permit. The runner contract's first version refused every request. Section 7 of docs/reference/harness-runner-contract.md now covers who is asked: With a default person, "They get an ordinary ask in the Feed and on Slack. No setup is needed." To choose someone else, run rig seat set-permissions-to <seat> --to <person> --reason "…". --clear returns to the default. When no one can be asked, the request is "Permitted at once, with the agent's own allow option." Refusal is opt-in, with --deny. "A person's no is never a grant." An unanswered request "stays pending and visible" until the agent's turn ends: "It's never timed out into a grant."
rig launch --seats first-starts seats; skill copies, activity hooks and startup declarations are kept across rig down, resumes and exports; and stops are clearer. A community contribution.Seats: an isolated tmux server per OpenRig home (rig config set terminal.tmux_server isolated); claude_mcp: none for a Claude member; resuming with a new token file; first starts on rig launch --seats; an accurate account record and kept effort after rig provider switch; skill copies, activity hooks and startup declarations kept across rig down, resumes and exports; clearer stops for handover's fork:<seat>, a missing conversation, claude not found, Codex restart refusals and failed unpins.
OPENRIG_TMUX_CONTROL=1 when control clients end together, and a submit key a read-only client blocks is pasted instead. A community contribution.tmux: with OPENRIG_TMUX_CONTROL=1, no crash on tmux 3.3a to 3.7c when the control client ends while another leaves; a submit key that a read-only client blocks is pasted instead.
Slack and usage: typed replies to parked decisions are recorded as answers; row bodies render Markdown; resolved decisions stop showing as live asks; OpenRouter charge retries aren't delayed.
rig terminal open mission:<id> --follow --provider herdr follows a mission. Linux Herdr 0.9.5 only. A community contribution.The TUI and views: agent pages show health and recent work; pod, spec and instance-table views read correctly; rig terminal open mission:<id> --follow --provider herdr follows a mission.
Restart the daemon after installing: run npm install -g @openrig/cli@0.7.8, then rig daemon stop && rig daemon start. Four migrations (142 to 145) store a Claude seat's MCP opt-out, seat activity grants, OpenShell environment bindings and the person chosen to answer a seat's permission requests.
The bundled openrig-core plugin advances from 0.1.23 to 0.1.24. One line of its seat-continuity skill changed. Managed refresh preserves replaced files under $OPENRIG_HOME/backups/managed-refresh/<id>/; reapply local edits you still need from that copy.
The install is larger: in the 5 November clean-install check, 0.7.8 resolves to 118 npm packages (including OpenRig itself), against 0.7.6's 32. The MCP SDK is now a runtime dependency.
A runner conformance check: node docs/reference/runner-conformance/check-runner.mjs --runner '<runner command with {agent}>' --name <harness>. It covers the answered path, and needs Node and tmux but no daemon, model, credentials or network.
dsh) comes with its preset to build against. OpenRig doesn't install it.An example runner for DeepSeek Harness (dsh), with its preset, is under docs/reference/community-harness-examples/dsh/. OpenRig doesn't install it.
A seat activity grant lets a runner outside the host's hook setup report activity for one seat and launch, without the install-wide hook token.
0.7.8 includes rig environment (setup, create, attach) and a reference for binding a seat to an NVIDIA OpenShell sandbox (daemon/assets/openshell/README.md). In our release testing the worker never became usable: after the sandbox was created and the native process started, rig environment connect failed when its check timed out after 5 seconds, so no team work ran. Don't rely on this path in 0.7.8. The packaged CHANGELOG and What's new describe it as working; this release body replaces them on that point.
rig host edit and rig host remove arrive, OpenRig warns when another user's daemon holds the port, and rig daemon status shows a pending start. A community contribution.Hosts and setup: rig host edit and rig host remove; a warning when another OS user's daemon holds the port; rig daemon status shows a pending start.
Wakes: OPENRIG_WAKE_SUSPEND matches destinations with a colon; prompt escalations stay deliverable without a reachable person; reminders stop at a full quota; human@kernel raises no false finding; OpenRig's own Codex hooks are trusted when no choice is recorded; refusals name their fix.
Also: MCP text is cut between whole characters; an ACP cancel survives a failed note; long transcript lines read faster; there are Pi and kernel dashboard recovery guides.
--deny keeps a seat refusing.A runner following section 7 gets answers or permits, not refusals. --deny keeps a seat refusing.
OPENRIG_TMUX_CONTROL=1 uses spawned clients instead of a control client.On tmux older than 3.7c, OPENRIG_TMUX_CONTROL=1 uses spawned clients instead of a control client.
OPENRIG_DSH and OPENRIG_DSH_HOME aren't accepted in a rig spec's env:. A preset command wrapper that exports them was the route we observed. (Fixed on main, #2414; not in 0.7.8.)/abort in the pane. A /abort sent with rig send is an ordinary prompt.allow_always from a permit: with no one to ask, "If the agent offers only allow_always, the harness then remembers a permission nobody chose."ANTHROPIC_API_KEY; subscription or OAuth credentials aren't supported. "Provider registration and image construction can incur costs."operator@kernel; the actual kernel operator is operator-agent@kernel.rig whoami --json ran on the host. The missing check is the same in 0.7.6; environments are a new way to trigger it.delete-owned didn't confirm its delete in our test, and the vendor's own delete cleaned up. A delete still finds the sandbox by name after reading its ID, so a deliberate name replacement in that gap is an open race.rig seat set-typing-guard <seat> --mode draft-aware protects an unsent draft.rig: command not found. In our first-user test, asking the agent to fix PATH recovered it.claude, a provider switch can't read the running effort (it uses the authored effort, and says so), and rig down may briefly run a resume probe.rig seat continue can refuse a healthy auto-mode Claude seat with attention_required. Ordinary sends still work.provenanceRecordWritten and departingSessionKilled as false even when both happened, so don't re-run a handover on those two fields alone. (Fixed on main, #2420.)Thank you to everyone whose work is in this release: @Aschenbath, @dajiaohuang, @Vaishnavi220506, @Nikhila-Dhevi-R, @Nikhi00718, @oodadoudou, @DeryFerd, @ch-davidriott, @vanducng, @Ethereal49, @AdityaxDeore and @Mydudeclaps; to @bpamiri for dsh-seat, whose fake agent the conformance check's stand-in follows; and, for reports carried into this release, @salbang, @ziadeh, @davey64, @bpamiri, @oneandonlydean, @xwcq, @seanmcn and @AlexGhanem.
Managed compaction no longer types into a draft. With the optional typing guard off (the default), compaction's preparation prompt could be pasted onto a message someone had typed but not sent. Now the preparation prompt, /compact and the restore prompts check the input box first: typed or unreadable input holds the step with a reason. A held restore prompt goes ahead on its own once the box is empty; a manual rig compact says why it stopped, to be retried once the box is empty. An empty box or Claude's own suggestion goes ahead.
A Claude seat under the "Update installed" line reads idle again. After Claude installs an update, it prints "✔ Update installed · /restart to apply" (or "Restart to apply") below its mode bar. OpenRig didn't recognize that line, so a seat at an empty prompt read unknown instead of idle, and managed compaction waited. Both spellings are now recognized; lines that only look similar still read unknown.
A Claude seat launched through a wrapper script can be rebuilt again. On 0.7.6, rig seat handover --source rebuild on a seat whose claude is a #! script that starts the real Claude without exec stopped after 30 seconds with "Waiting to confirm the new Claude process in this pane", though the new Claude was running. The readiness check now also accepts the script's own direct child when it names this launch's conversation; every other check is unchanged.
Restart the daemon after installing so the fixes take effect: run npm install -g @openrig/cli@0.7.7, then rig daemon stop && rig daemon start. There are no database migrations. The bundled openrig-core plugin stays at 0.1.23 with unchanged files.
A failed handover no longer moves the seat's resume target to the new, unprimed conversation while the seat keeps its previous occupant.
Thank you to @oneandonlydean for the report and the evidence in #2341.
rig provider switch moves a running agent to another account home, with a fresh conversation and a handover; it keeps its address and work. A community contribution.Move a running seat to another account home with rig provider switch --seat <seat> --to-account <name> --account-home <dir> (preview with --dry-run). It starts a fresh conversation with a handover, and the seat keeps its address and work.
rig terminal open slice:<name-or-id> --follow --provider herdr shows the agents with open work in that slice, each once. Linux Herdr 0.9.5 only.rig terminal open slice:<name-or-id> --follow --provider herdr shows the current owners of the slice's open work: pending, in-progress and blocked queue rows tagged with that slice. Each owner appears once, with all its obligations listed. Tag a row with rig queue create --slice <name-or-id>; handoffs keep the tag.
Slice following targets Linux Herdr 0.9.5 / protocol 22. Following a whole mission isn't in this release.
rig daemon start says startup is pending and leaves the daemon running, instead of killing it.Startup: when only the readiness deadline passes and the daemon is still starting, rig daemon start exits nonzero with a pending message naming the process, its endpoint and the log. The daemon is left running, and a second start stays blocked. rig daemon stop can't reach that unpublished process; the message gives the manual restart.
A queue wake refused at a selection prompt is retried once the prompt clears. Work sent to a terminal-only seat raises a finding for a person. Cross-host rig queue create --verify reads the destination's receipt.
Restart the daemon after installing: run npm install -g @openrig/cli@0.7.6, then rig daemon stop && rig daemon start. Two database updates record a seat's selected account home and whether a community-harness launch declared a runner.
The bundled openrig-core plugin goes to 0.1.23: its Claude activity hook now clears permission marks after a tool batch. Replaced files are preserved under $OPENRIG_HOME/backups/managed-refresh/<id>/; reapply any local edits you still need.
rig daemon start may report that startup is still pending. Follow the guidance it prints before any recovery.On a large database, rig daemon start may now report that startup is still pending instead of failing. Follow the guidance it prints before trying any recovery.
New slice views start with four automatic tiles; --max-tiles changes that. A known empty slice keeps following and fills again. An incomplete read keeps the previous viewers, and missing or ambiguous slices are named, never shown as an empty team. Pins, drafts, unknown input and focus still protect a departed viewer, and closing a viewer never stops its agent.
When a slice's folder name repeats across missions, tag its work with the slice's unique ID (or add --mission). A row tagged only with the shared name can't be placed, so the view reads incomplete until it is.
Slice following targets Linux Herdr 0.9.5 / protocol 22. Following a whole mission isn't in this release.
An owner whose join was refused before any pane split now joins on a later refresh, and a viewer that hasn't attached yet can carry on.
Named Herdr sessions use their configured endpoint throughout, and the TUI's terminal action opens the agent, pod or rig in view.
A Claude seat can switch between subscription and API key on the same conversation (--to-account api --api-key-file <file>, back with --to-account plan --token-file <file>). In-flight work is interrupted. The API-key path is not qualified yet, so check the resulting account and conversation. A new Claude member can name its own token file in its spec with claude_token.
Managed seats inherit the daemon's SSH agent for git over SSH.
Community-harness presets can opt into runner: true (with launch_result: true) so the seat's activity follows its runner, and passes_model: true to receive the declared model as OPENRIG_MODEL. That says which model was passed, not which one runs. New draft author contracts describe harness, runner and platform duties, marking what's implemented and what's proposed.
rig expand keeps the member settings rig up accepts, and a restore keeps a model set with rig seat set-model.rig expand keeps the member settings rig up accepts, including env, effort and startup settings, and a restore keeps a model set with rig seat set-model.
view_required and unknown_provider, where it returned 200 with an error body.Viewer API: POST /api/terminal/viewer/ensure returns HTTP 400 for view_required and unknown_provider, where it returned 200 with an error body. Keep reading the body's code.
A verified exited process in a retained pane reads pane_exited, not running and idle, and its work goes to the usual recovery finding.
Smaller fixes cover inventory reads, UTC snapshot ages, TUI emoji, help text and test fixtures.
rig status then shows the kernel as partial_ready, and rig ps shows the seats needing attention. No message is sent. This also happens on 0.7.5.humanAnswers doesn't hold for such a parked question.rig seat held-messages before resending. A larger pane or a shorter message may help; turning the guard off removes draft protection and isn't a fix.rig seat restart <seat> --keep-conversation refuses, nothing is stopped; leave the seat running and report it with your CLI and daemon versions.Thank you to everyone whose work is in this release: @seanmcn, @DeryFerd, @Aschenbath, @Vaishnavi220506, @bradvin, @jay23606, @rudycelekli and @Nikhi00718; to @akushonkamen, @jcmurrayii, @KM199 and @pdurlej for changes, reports and reproductions that shaped it; to @appleguy, @al-robertson-cto and @skv-io, whose reports led to fixes and documentation here; and to @bpamiri, whose runner-preset proposal on #43 shaped the runner field.
rig terminal open qitem:<id> --follow --provider herdr keeps the agents working on a task together in one Herdr workspace.rig terminal open qitem:<id> --follow --provider herdr shows the current owners of a task's open work as viewer tiles in one Herdr workspace. Joining agents get viewer tiles; departures close viewers, not agents or conversations. --pin, --suppress and --max-tiles adjust the view; --no-follow keeps a fixed layout. Defaults are eight tiles, four seconds of departure grace and twelve more seconds for a focused leaver.
The preview is limited to Linux Herdr 0.9.5 / protocol 22 and requires no internet exposure.
rig send waits up to 60 seconds for its result. An unconfirmed send may still have arrived, so check before sending again.rig send waits up to 60 seconds for its result, locally and over HTTP, plus a requested idle wait. A send can still read unconfirmed sooner, for example when a connection drops. An unconfirmed send may still have been delivered; input may be held off-screen by a busy recipient. Reconcile before resending. No automatic resend is added.
With a Codex seat's typing guard set to draft-aware or openrig-urgent, a message OpenRig can't confirm will stay visible in the composer is held before it's typed, so nothing is left typed and unsent. A message that fits is delivered and submitted once, and a person's draft stays protected. The default (guard off) and other runtimes are unchanged. Delivery's guards also recognize Codex 0.164's two-row composer footer, so a truly empty Codex 0.164 composer can receive typed delivery.
A wake recorded as unroutable is retried through its existing intent once the seat has a pane; stuck findings require delivery or custody evidence to close.
rig down after a rolled-back launch stays down. A community contribution.Fresh Claude handovers wait for the new process and its current input frame. A rolled-back launch followed by rig down stays down on rig start --last.
escalates_to edge that did nothing before now takes part in recovery for an agent with no delegation parent (no agent that hands it work).Restart the daemon after installing: run npm install -g @openrig/cli@0.7.5, then rig daemon stop && rig daemon start. A database update stores deferred escalation addresses while preserving existing edges. Previously inert escalates_to edges now affect recovery for seats without a delegation parent.
Startup refreshes openrig-core to 0.1.22. Its queue-handoff skill teaches agents to park long waits with a continuation and wake. Replaced files are preserved under $OPENRIG_HOME/backups/managed-refresh/; reapply local edits you still need. No setting is added or removed.
If Codex requests hook review, review the hooks you intend to enable. The earlier hook-refresh remedy, rig seat restart <seat> --keep-conversation --reason "refresh hooks" (preview with --plan), has an unresolved refusal report; read #2200 below before relying on it. Upgrading does not restore approvals an earlier OpenRig removed.
escalates_to address in another rig on the same daemon. A community contribution.Parentless recovery can use an escalates_to address in another rig on the same daemon. A delegation parent still wins; an open recovery keeps its recorded owner.
rig reconcile-session as the fix.Runtime send refusals return HTTP 409 instead of 500. A stale Claude pane names rig reconcile-session <seat> as the repair.
Follow mode retains departed tiles with drafts or unreadable input. A completely empty task drains removable tiles and stops following once none remain; run --follow again to resume that workspace.
The Overview reads the selected rig's current configuration; the TUI uses the configured local host name; snapshot status matches the scheduler.
Terminal results distinguish pane_unconfirmed from opened and skipped panes. An all-unconfirmed open exits nonzero; inspect before retrying because something may have opened.
pi-tui example, clearer hosted-member guidance (hosted members stay experimental, unsupported, off by default and at your own risk), and tmux, capture, port and scope-folder repairs. A community contribution.The pi-tui example runs Pi's own terminal UI with a per-seat agent folder and startup input. Hosted-member guidance says how a member without a wake trigger gets work, that a listed tool is not a grant, and that finished work closes with no-follow-on. Other changes repair tmux separator parsing for attach --self replies, normalize capture/transcript line counts, warn on invalid MCP daemon ports, release unread notification response bodies, and accept scope folders numbered 100 and above.
Hosted members retain 0.7.4's framing: experimental, unsupported, off by default and at your own risk. Local assistants can connect locally. Internet access requires your own TLS tunnel or proxy and a narrowly scoped token; anyone holding a valid token can act within its grant. Revoke it when finished with rig mcp token revoke <name>.
What you'll usually see is one agent. A queue item has one owner at a time, and handing it off closes it, so following an ordinary task shows its current owner, replaced by the next owner after a handoff. Several agents appear together only when several open rows descend from the task at the same time, by handoff or by naming it in their chain of record.
Clients can add or reuse a verified viewer without rearranging existing panes through POST /api/terminal/viewer/ensure. Membership grants no input authority.
Following larger scopes of work, such as a slice or a mission, isn't in this release.
The view recognizes Codex 0.164's two-row composer footer, so a departed Codex agent's empty composer no longer reads as unconfirmed input that keeps its tile.
draft-aware or openrig-urgent, OpenRig holds a message before pasting it when it can't confirm the whole message will stay visible in the composer. That can happen in small task-following tiles, and many wrapped messages are held even when they would have fit. A held message is not delivered. Use a larger pane; a shorter message may also fit; turning the guard off removes draft protection and is not a fix. A very long message (over 1,000 characters) is held whatever the pane size, so shorten it.--session <name>, OpenRig may resolve Herdr's default socket location rather than a custom endpoint your daemon is configured to use. The configured endpoint works without --session. To reach a specific endpoint, set HERDR_SOCKET_PATH to it explicitly; don't rely on XDG_CONFIG_HOME, which the default resolver doesn't read.ignore-size is not isolation.pi-tui unattended, without typing into it: a wake can submit a draft someone is typing or enter an unrecognized dialog. Busy-turn wakes and native resume remain unqualified; a plain terminal receives no queue wakes. See the example's limits.rig down stops a community harness's pane; inspect any independently spawned processes before relying on their exit.Thank you to everyone whose work is in this release: @icey-zzz, @Aschenbath, @DeryFerd, @Nikhi00718 and @dajiaohuang; to @rudycelekli for the work completed in #2160 and @todie for the snapshot-status change followed in #2168; and, for reports, @agraubert, @AdamWhitehurst (#1544, #2049), @kevinhsykes (#2151), @aanestesia (#605), @todie (#1419) and @tgrundtvig (#21).
Task following was checked with real agents on Linux Herdr 0.9.5: tiles joined and left as genuine work moved (3 → 4 → 3), a focused leaver stayed briefly before closing, a person's typed draft survived a join, the tile cap, pins and suppression behaved as described, and turning following off and on again kept the same agents. The empty-task stop was confirmed from the recorded events, and on screen in a recorded run. Resource use was measured only as whole-machine CPU in one window, not for the daemon alone.
Performance against 0.7.4 was measured on an earlier build of this release, on one machine, with a synthetic workload: no measured change in command latency, memory or CPU.
Protected Codex delivery (#2242) was checked with real Codex agents: a fitting message at 80×24 was submitted once (a model reply was not checked), in a 30×9 pane it was held before paste in both guard modes, and at 32×9 a person's unsent draft was kept in both modes. The 65×9 tile, messages with hard line breaks, and very long messages were not run natively.
Assistants on your machine can already reach OpenRig's MCP server locally. A cloud-hosted assistant can join a rig over MCP under a grant for its seat. Its roster entry declares "hosted": true; rig_team shows its capabilities, caveat and activity as unknown, never idle. The connection teaches it to check its inbox, claim work, return an answer or hand it on, and keep work that needs a person's approval until the answer arrives. This is experimental and unsupported. It's off by default. Nothing in OpenRig listens on the internet unless you set it up yourself. You serve rig mcp serve --http, put it behind your own TLS tunnel or proxy, and issue a token. Doing that is like opening a port in your firewall. Anyone with the URL and a valid token can act as that member within its grant. Treat the token like a password, give the member the narrowest grant, and revoke it when you're done (rig mcp token revoke <name>). We don't support, and take no responsibility for, an OpenRig reachable from the internet, just as with running OpenRig itself on the open internet. Use at your own risk. A real Grok claimed work, returned a useful note, came back in a new conversation and was refused after its grant was revoked, with its work kept. Other assistants weren't tried.
Resume the exact Claude Code or Codex conversation you choose. rig seat history --rig <id> --node-id <id> lists the Claude and Codex conversations that exact node could resume. A client can then resume the history a person reviewed. It refuses rather than substitute another conversation or start fresh. Resuming a reviewed history has no CLI yet. It works only through the daemon's startup review, for clients. With Codex 0.164 on a low-effort model, the resumed conversation didn't recall an earlier fact, and the same happened outside OpenRig. This completes the three stages agreed in #1837.
Clients on MCP protocol revision 2026-07-28, which has no initialize handshake, now work with rig mcp serve over stdio or --http. Existing 2025 clients still work on the same endpoint. A 2026-07-28 client (SDK 2.x) was shown working at installed scope. Serving over --http from the internet is experimental, unsupported and off by default. Anyone with the URL and a valid token can act within its grant. Give the narrowest grant, revoke it when you're done, and use it at your own risk.
rig terminal open and the daemon's terminal preview accept qitem:<id> as a read-only membership view. It names stopped, missing and external owners. It doesn't follow a task as it moves, keep a reusable viewer or grant input. An explicit open still has the provider's usual workspace effects.
Parks on a registered person's own address resolve only from a Slack reply written after the question. They keep their summary and evidence and show in NEEDS YOU.
Install with npm install -g @openrig/cli@0.7.4, then run rig daemon stop && rig daemon start. On first start, a database update adds a column recording which grant's caller acted. Startup refreshes openrig-core 0.1.21: the queue-handoff skill and Claude hooks changed. It first copies replaced files under $OPENRIG_HOME/backups/managed-refresh/; reapply any edits from the copy. Claude seats get the new rate-limit hook at their next launch, resume, restore or handover. If a Codex seat stops at "Hooks need review", review the hooks in Codex. A Codex seat running through the upgrade warns "Hook failed … plugin contents changed; hook approval required", and some OpenRig hooks stop running in it. Codex 0.164's /permissions menu has no hook item. Preview with --plan, then restart the seat on its own conversation with rig seat restart <seat> --keep-conversation --reason "<why>": the warning goes away and OpenRig's hooks run again. This was tested on Linux with Codex 0.164; seats started after the upgrade aren't affected. Going back to 0.7.3 leaves the new column in place, unread.
A hosted member's roster entry can declare a webhook trigger using secret names only. OpenRig sends a pointer to the work when it lands. A 2xx response means the endpoint accepted the ring, not that the member woke. If it doesn't claim the work in time, its person is told. rig_team and rig roster show show the member's last MCP call; rig_team also shows its latest ring. No webhook consumer was shown waking from a ring. An ordinary chat app cannot receive unsolicited events, so without a webhook-capable runner the assistant sees work only when it checks. Hosted members and internet-facing MCP are experimental, unsupported, off by default and used at your own risk. Only Grok was tried.
When work waits for a member nothing can wake, OpenRig tells its person on Steering, the TUI's Feed and Slack. The notice goes to the person, not the member. For hosted members, this notice wasn't shown with a registered person. Hosted members and internet-facing MCP are experimental, unsupported, off by default and used at your own risk. Only Grok was tried.
Rows record whose grant acted: attribution, not authentication. A caller's seat that was created but never launched is served instead of refused. A caller granted through a grant service hands work to its granted team, never to a caller seat, and cannot attach tags that hide text from a person asked about it. For advanced remote setups, rig mcp serve --http --grant-service <url> checks each request's key with an external grant service, with nothing cached. --oauth-resource and --oauth-issuer add OAuth protected-resource discovery (RFC 9728) and standard sign-in challenges; the key and grant still decide what a caller may do. This is experimental and unsupported. It's off by default. Nothing in OpenRig listens on the internet unless you set it up yourself. You serve rig mcp serve --http, put it behind your own TLS tunnel or proxy, and issue a token. Doing that is like opening a port in your firewall. Anyone with the URL and a valid token can act as that member within its grant. Treat the token like a password, give the member the narrowest grant, and revoke it when you're done (rig mcp token revoke <name>). We don't support, and take no responsibility for, an OpenRig reachable from the internet, just as with running OpenRig itself on the open internet. Use at your own risk.
A Claude seat stopped by a rate limit is recognised from Claude's StopFailure hook, and a delayed older hook no longer replaces a seat's current state (#1908, #2113). A fresh Claude handover reports delivery only once the packet is submitted (#1863). Periodic snapshots keep a down agent's resume source (#2123). A resumed Pi seat no longer prints Entry not found on every restart (#2122).
The TUI starts in a terminal that reports 0 rows or columns (#2099) and lists rigs by name (#2134). The reviews count only open rows as held (#2136). Identity lookup no longer prints tmux errors (#2103). The packaged CHANGELOG.md lists every change.
Startup brings back a kernel only when this start found it down, which is what a reboot leaves. A kernel already down before the start stays down. rig start and rig daemon start --wait-for-kernel explain why and show rig up kernel --existing as the way back. The wait-for-kernel command still exits 1 for a deliberately skipped kernel, but no longer reports a timeout and an authentication fix. This fixes 0.7.3's known issue of a long-gone kernel returning. Restore evidence lasts one daemon process: if the first daemon after a reboot exits before restore starts a seat, the next start leaves the kernel down and says so.
rig up kernel --existing.rig export doesn't write a README, and openrig.dev/rigs' check fails until one is added and declared in rig.yaml docs. Add it, then run rig bundle check on the folder before you submit.rig down stops the pane only.ps before relying on it.rig attach --self can send, but replies to it may fail with transport_unavailable.unconfirmed. The packet can be left typed but not sent in a Claude successor's prompt. In our test, the seat's next ordinary task got it consumed.Thank you to everyone whose work is in this release: @Aschenbath, @Nikhi00718, @bradvin, @dajiaohuang and @seanmcn, and, for reports and reviews, @pr0xc3nt4ur1 (#1822), @duncanleung (#2130), @ziadeh (#2129), @alperta (#99), @AdamWhitehurst (#1544), @smota (discussion #2117) and @oSquashBlossomo.
Register more than one person (rig gateway human add) and choose the default (rig gateway default <id>), who receives the last escalation step, the notice digest and workflow fallbacks.
rig gateway channel add <id> --binding slack:… registers a Slack shared channel as one participant. Anyone writing there reaches the team, and each row names the channel, the sender and the policy. A channel's messages never answer or close an ask.
rig gateway participants shows where it came from. It informs their judgment; nothing enforces it.rig gateway org set and group set give organizations and groups a readable policy, and rig gateway participants shows it with its source. It's context for your agents' judgment; nothing enforces it.
rig seat set-permissions <seat> --codex-permissions <file> gives a Codex seat its own sandbox and approval settings on every managed launch.rig seat set-permissions <seat> --codex-permissions <file> gives a Codex seat its own sandbox and approval settings, passed on every managed launch.
docs/reference/community-harness-presets.md is the contract: what a harness gets from OpenRig and what it doesn't, how to write one, and how promotion is judged. Tests cover it.
Restart the daemon after installing: run npm install -g @openrig/cli@0.7.3, then rig daemon stop && rig daemon start. One database update applies on the daemon's first start. Startup refreshes OpenRig's own plugin (openrig-core 0.1.20; the messaging-the-human skill changed), first copying any file it replaces under $OPENRIG_HOME/backups/managed-refresh/. If you edited one, reapply your change from the copy.
rig gateway default <id>. Until you do, the last escalation step and the notice digest wait.If you register a second person, choose a default: rig gateway default <id>. Until you do, the last escalation step and the notice digest wait.
ANTHROPIC_BASE_URL, declare it in the rig's or member's env:.If a seat relied on the daemon's own environment (ANTHROPIC_BASE_URL, OPENAI_BASE_URL, CLAUDE_CODE_USE_BEDROCK), declare it in the rig's or member's env:. The first seat on a new tmux server now gets the same scoped environment as every other seat.
Going back to 0.7.2 keeps every registered person, but not the default person, shared channels, organizations, groups or policy. It keeps a Codex seat's permission settings without reading them, so its launches of that seat can be broader.
rig down names community harness seats whose processes may outlive their pane, and a Hermes gateway runs per seat.An OpenClaw seat whose last local turn was refused shows as needing attention, rig down names community harness seats whose processes may outlive their pane, and a Hermes gateway runs per seat with the example launcher.
A registered person can't settle an ask that was asked of someone else (answer_not_recorded); from the TUI, the primary user can.
target_needs_input.A send to a Claude seat at its "Resume session" picker refuses with target_needs_input.
rig seat set-permissions <seat> --codex-profile <name> picks which Codex configuration profile a seat uses on its future managed launches. Requested by @bpamiri.rig seat set-permissions <seat> --codex-profile <name> selects that seat's Codex configuration profile for future managed launches. Requested in #1534 by @bpamiri.
Codex's own trust and config edits survive OpenRig's next launch, and a single-quoted trust table no longer gains a duplicate that stopped Codex starting. Resume handles a path with a space, and a seat started under a launcher gets its generation record.
rig slack post-open-asks posts the asks still open, and the reply hint on a person's message now answers in their Slack thread.rig slack post-open-asks posts the asks still open, and the reply hint on a person's message now answers in their Slack thread.
When an agent parks its own ask on the person it asked, a colleague's reply in that thread goes to the agent. Incoming Slack messages are not automatically reposted to Slack as new asks; an agent can still deliberately ask a follow-up.
whose filter chooses whose queue you see, and settling someone else's ask from the TUI is recorded as the primary user's, with the person asked kept.The Feed's whose <person|all|default> chooses whose queue you see. Settling someone else's ask from the TUI is recorded as the primary user's, keeping the person asked: attribution on a trusted network, not a login.
The TUI opens views in cmux when it runs inside cmux, and keeps a registered remote host in its tree on every tab.
rig capture --readable reads a small seat at a readable size, then gives its window back.rig capture --readable reads a small seat at a readable size and gives its window back, and a window gets its own size back after the web terminal.
rig doctor checks that tmux can capture a pane, and new seats keep Claude's Shift+Enter.rig doctor checks that tmux can capture a pane, and new seats keep tmux's TERM_PROGRAM for Claude's Shift+Enter.
rig scope --workspace <path> mission create refuses a path that doesn't exist.rig scope --workspace <path> mission create refuses a path that doesn't exist.
--wait-for-idle waits work, a dropped local send reads unconfirmed, rig export keeps pod startup files, and the delivery digest says why it sent nothing. A community contribution.Also: --wait-for-idle over 300 seconds works; a dropped local send reads unconfirmed; saved seat settings can target a node id; rig export keeps pod startup files; rig scope --workspace creates the first mission there; the delivery digest says why it sent nothing.
rig down stops the pane only; check the harness's own processes afterwards.ps before relying on it.rig or rig tui started in a terminal that reports 0 rows or columns crashes on its first frame, as in 0.7.2. Start it in a terminal with a size. The fix (#2099) is on main for the next release.rig down kernel; a kernel stopped that way stays down on later starts. rig daemon start --no-kernel skips kernel startup entirely. A fix is planned for a later release.communityHarness) and its original database update; renaming them would break a return to 0.7.2.Thank you to everyone whose work is in this release: @Aschenbath, @Coder8124, @bradvin, @sn1thy and @DeryFerd, and, for reports and diagnoses, @AdamWhitehurst (#2041, #2049, #2057), @borfast (#2029, #17), @jimallen (#17), @R14014 (#1069), @kainne44 (#70), @VltrnOne (#1653), @d3fvxl (#800), @fsafey (#1799), @MichelMabinuola (#980), @rakoort (#1679), @bpamiri (#1534) and @george661 (#1710).
rig seat set-runtime <seat> --runtime claude-code|codex moves a live seat between Codex and Claude Code, keeping its address, its queued tasks and its claims. A community contribution.rig seat set-runtime <seat> --runtime claude-code|codex --reason "<why>" moves a live seat between Codex and Claude Code. The successor starts first, gets a handover and confirms it before the old runtime stops, and the seat keeps its address, queue rows and claims. --dry-run shows the settings, what is cleared and the costs; --recover finishes an interrupted switch. Built on @Aschenbath's design in #1214.
env:.A seat can run on its own Claude or Codex login. Sign in to a separate home once, natively, then name it as an absolute CLAUDE_CONFIG_DIR or CODEX_HOME in the member's or rig's env:. It applies at the seat's next launch, and rig seat status and rig provider status show which home a seat uses and where it comes from.
rig seat set-permissions <seat> --rules <file.json>.Claude tool rules per member. Claude members that share a project folder can keep their own tool rules: claude_permissions with allow and deny lists in the spec, or rig seat set-permissions <seat> --rules <file.json> --reason "<why>" for an existing seat. A change is audited and applies at the next managed launch or resume; OpenRig never writes the rules into the project's Claude settings. Checked natively on Claude Code 2.1.309: two seats in one folder enforce their own rules, a managed policy wins, and fresh, resume and fork launches enforce them. See Known issues for what isn't claimed.
A working Claude seat that quotes approval-prompt text in its own output, such as › 1. Yes, proceed (y), no longer has sends and queue wakes refused as if it were at a prompt, including when Claude draws its input box high in a tall pane. Real permission, question and MCP dialogs still refuse.
LEARNED.md.A Codex seat on the default workspace-write sandbox can still commit after a resume, fork or restore, and can write its own topology folder, so it keeps its LEARNED.md. When the sandbox refuses its recap, the daemon saves it.
Restart the daemon after installing. Six database updates apply on its first start: run npm install -g @openrig/cli@0.7.2, then rig daemon stop && rig daemon start.
Startup replaces OpenRig's own plugin and skill files (openrig-core goes from 0.1.18 to 0.1.19; the delegating-work and openrig-user skills changed). As in 0.7.1, it first copies each file it replaces under $OPENRIG_HOME/backups/managed-refresh/ and logs where, edited or not. If you did edit one, reapply your change from the copy.
env:.On Linux, if a seat's tools use the session bus, set the bus address in the rig's or member's env:. Seats now start with an address that can't connect, even when a shell startup file sets XDG_RUNTIME_DIR (see Behaviour that changed and Known issues).
CLAUDE_CONFIG_DIR or CODEX_HOME in the rig spec's env:. A one-off rig up --env no longer applies them.Declare a seat's own CLAUDE_CONFIG_DIR or CODEX_HOME in the rig spec's env:. A one-off rig up --env no longer applies those two, and says so.
If you might go back to an earlier version, set equivalent native Claude rules first. Code from before #1914 opens the upgraded database but leaves per-member Claude tool rules out of launch, so a seat's denies stop applying. docs/reference/rig-spec.md has the details.
rig seat clear-attention <seat> clears it, and the send refusal says so.After a person declines a Claude permission prompt, rig seat clear-attention <seat> --reason "<what you saw>" clears the prompt the latest hook still names, and the send refusal says to.
rig launch <rig> <seat> first-starts one seat of a team imported with --materialize-only.rig launch <rig> <seat> first-starts one seat of a team imported with --materialize-only, without restoring the others.
On Linux, a seat launched by OpenRig now gets the daemon's XDG_RUNTIME_DIR, with DBUS_SESSION_BUS_ADDRESS=disabled:, a session bus address that can't connect. Tools whose sockets live in the runtime directory work in seats without per-tool setup, and a seat's D-Bus clients don't find or start the session bus on their own. A seat's own XDG_RUNTIME_DIR, from env: for example, leaves both as before, and its own bus address is kept. Unchanged: macOS, a daemon started without XDG_RUNTIME_DIR, and terminal.inherit_tmux_environment on. See Known issues.
/command typed mid-turn runs at once instead of being refused.In a Pi seat that OpenRig runs, an extension /command typed mid-turn runs at once instead of being refused as a steer.
A pod seat that never launched starts fresh from its saved launch recipe after a daemon restart, instead of waiting at a shell.
rig down settles its seats, so rig archive and rig up can go ahead.After a team's tmux server dies, rig down settles its seats, so rig archive and rig up can proceed.
rig start, and rig down names the terminals it stops.A team of terminal seats reads fully restored after rig start, and rig down names the terminals it stops.
A launch whose tmux made no session fails as session_not_created or session_unconfirmed instead of reading launched.
A seat held at a choice prompt that no hook announces, such as Claude's usage-limit dialog, now escalates while it holds work.
A Pi or OMP seat's second startup block reaches the running turn instead of raising runtime_error.
rig stream act <id> records what a reader did about a stream item, and rig stream show lists it.rig stream act <id> records what a reader did about a stream item: --work <row>, --decline "<reason>" or --duplicate <id>. rig stream show lists each reader's current action.
rig send to a Codex seat mid-turn says the message joins the running turn, and names --wait-for-idle for separate work.rig send to a mid-turn Codex seat says the message joins the running turn and may redirect it, and names --wait-for-idle for separate work.
rig send to a seat switching runtime is held, not typed; send it again after the switch.A rig send to a seat that is switching runtime is held, not typed (outcome: retained, reason: runtime_switch_in_progress); send it again after the switch.
stuck-sweep@system. A community contribution.A row OpenRig creates itself, such as a stuck-sweep finding or a workflow escalation, wakes its seat from OpenRig's own sender with no reply line, instead of from the row's creator. A stuck-sweep closure is recorded as stuck-sweep@system.
agents.operator_session set, a dead seat's finding goes to that agent first, then the operator seat.With agents.operator_session set, a dead seat's finding goes to that agent first, then the operator seat (workspace.operator_seat_name), then the row's creator. Approvals still reach the person. With it unset, routing is as before.
queue.wake_human_rung, and only one reaches a person at a time.An alert for a seat at a choice prompt follows queue.wake_human_rung: under explicit-only it stays with the operator seat as a visible row, unless that seat is a person. Only one choice-prompt alert reaches a person at a time on a machine; the others wait as visible rows and post when it retires. Permission-prompt alerts post as before.
Cross-host queue writes wait 15 seconds for the daemon's answer, and a timeout reads "queue request got no answer in time" with the ID to reconcile by.
A draft in a Claude seat's input box that starts with a blank line no longer reads as an empty box, so a send isn't typed into it.
A draft-aware send to a Claude seat keeps ownership of its paste when Claude's footer changes to a hint such as "paste again to expand" (see Known issues for #14).
An ordinary rig send to an adopted Pi window waiting at a confirmation dialog is refused with target_needs_input and types nothing, instead of choosing the highlighted answer (#2015). A later change extends that recognition to Pi's input and editor dialogs, including help rows that wrap in an 80-column pane (#2026). A Pi seat that OpenRig runs shows an extension dialog in its pane, cancels it and marks the seat as needing attention (#1996).
A queue wake refused at an approval prompt is retried once when the seat's attention clears, instead of waiting for the stuck sweep.
rig queue resolve <qitemId> --decision "<evidence>" also releases a task held until a typed blocker such as external: clears. A community contribution.rig queue resolve <row> --decision "<evidence>" also releases a row parked on a typed blocker such as external:, fold: or auth:.
rig queue whoami and rig proof add find work in other catalogued projects, and rig scope mission create writes the mission's name. A community contribution.rig queue whoami resolves a project-tagged row against its catalogued project, rig proof add finds a slice in another catalogued project, and rig scope mission create writes the mission's name.
rig queue create whose outcome is unknown reads the task back once before reporting, and a timed-out cross-host write reads INDETERMINATE.A rig queue create whose outcome is unknown reads the row back once by its ID before reporting, and a cross-host write that times out reads INDETERMINATE instead of failed.
A dead seat's stuck-sweep finding goes to someone who can act, not to the dead seat, and says which rule chose them.
The parked-owner reminder no longer reaches a seat in the first 5 minutes after it starts waiting, such as just after a handoff or a question to a person. A seat at an approval or question prompt gets no grace.
A stalled-after-claim finding's age counts from the stall, not the claim.
haiku with the plan permission is told that Claude Code answers its plan-mode turns with Sonnet, and how to set a full Haiku id.A Claude seat set to the alias haiku with the plan permission is told at launch, at restore and in rig seat status that Claude Code answers its plan-mode turns with Sonnet, and how to set a full Haiku id.
When Claude refuses a startup turn late, for example on an expired login, OpenRig checks again before the next startup input's Enter and holds the rest as owed.
The model-divergence monitor no longer flags a Claude seat pinned to an undated id that answers with its dated snapshot, or to a Bedrock inference-profile id.
hooks.json, OpenRig's activity hooks go there instead of config.toml.On a Codex home with a hooks.json, OpenRig's activity hooks go there, and its own old inline block moves out of config.toml.
rig seat restart --keep-conversation on a Codex seat handles a path with a space, finds the resumed conversation, and says honestly when it can't prove the resume.rig seat restart --keep-conversation on a Codex seat keeps a writable root whose path holds a space, looks for the resumed conversation in the home the running Codex uses, and, when it can't prove the resume, says the pane restarted instead of "nothing was restarted".
hooks.json, OpenRig puts its activity hooks there. Your own hooks, their trust and your disables stay. A community contribution.When a Codex home has a hooks.json, OpenRig puts its activity hooks there and moves its own old inline block out of config.toml. Your hooks, their trust and your disables stay; @fsafey found the mixed-format problem in their fork. Inline hook trust is keyed by the managed group's actual position.
login_required instead of waiting out the readiness timeout. A community contribution.A signed-out Codex seat at the sign-in chooser or API-key entry reports login_required instead of waiting out the readiness timeout (Codex 0.164.0 and 0.162.1).
From <seat> on <machine>, so update anything that read the sender from the last line.Every Slack post starts with From <seat> on <machine>, and an escalation's headline reads Waiting on <seat>: <summary>; update anything that read the sender from the last line.
Slack: every post names its sender first, an escalation names the seat the work waits on, a reply from another registered person reaches the agent as their own message instead of answering someone else's question, and Slack rows name who wrote.
rig capture says when a pane smaller than 80x24 may cut off a seat's prompt, with the command to resize it.rig capture says when a pane smaller than 80x24 may cut off the prompt a seat is at, with the command to resize it.
OpenRig's MCP server adds rig_queue_handoff and rig_queue_unclaim (36 tools), so an assistant can pass held work on, or put it back, in one step.
rig up --env CLAUDE_CONFIG_DIR=… or CODEX_HOME=… isn't applied, and the CLI says so. Name a seat's home in the spec.rig up --env CLAUDE_CONFIG_DIR=… or CODEX_HOME=… isn't applied, and the CLI says so; other --env values work as before. Name a seat's home in the spec's env:.
rig seat status --json always carries permissions.account.advisories.rig seat status --json always carries permissions.account.advisories (empty unless there's an advisory).
rig acp run --help names now ships: rig context get reference/acp-agents.md.The guide that rig acp run --help names now ships: rig context get reference/acp-agents.md.
From contributors: queued terminal input is dropped after its client disconnects (#1636, @Vaishnavi220506); TUI focused summaries load their health findings (#1986, @Kero-66 and @Aschenbath); the TUI keeps the stream's connection state across pages (#2014, @Nikhi00718); the web dashboard reads a skipped kernel boot as "Boot not tracked", not "Unknown" (#1917, @Aschenbath, reported by @Kero-66); the compaction-and-restore scenario runs through a launch script, and the test-system docs state the CI job's real counts (#1924, #1925, @DeryFerd).
Guidance: the installer guide says what setup and starting Codex seats write to your home configuration, and the operator puts project work on a team's queue; a Codex user who declines the remembered command rule is offered the operator conversation; rig queue update --help and its refusal say how to cancel a row (--state done --closure-reason canceled); rig skill audit says what its stale and hash findings mean; 14 subcommands that showed a blank help summary now have one; the openrig-user skill says a queue create's request ID isn't the row; delegating-work also applies when judging a delegated result; and docs/reference/seat-delivery.md lists all six typing-guard modes.
--plan warns that it would.A service bundle install on a machine without Docker Compose v2 stops with compose_missing before it writes the target or creates the team, and --plan warns that it would.
rig export copies the team's culture file and names each seat whose agent ref points outside the export.rig export copies the team's culture file and names each seat whose agent ref points outside the exported folder.
rig compact <seat> --state shows a restore that is still owed, and repeated failures open one queue item for whoever can act.rig compact <seat> --state and seat status show an owed restore stage and its last delivery result, and repeated failures open one queue item for whoever can act.
RECAP.md or recap-superseded/.The recap store refuses to write through a link at RECAP.md or recap-superseded/ (recap_write_refused), a layout that used to work.
A managed Claude restore reads its own restore files without stopping at a Read approval, including on macOS, through a symlinked home or project, and for the generic or configured post-compact-extra.md and preparation root. The grants apply from the seat's next launch or resume.
A seat restored with claude --resume that stops at Claude's resume chooser in a tall pane is reported as waiting for a person's choice instead of resumed, and a Codex model chooser there reads as model_selection_gate. An old chooser above a resumed session no longer reads as live.
After a managed compaction, a Claude seat's restore starts on the next poll instead of waiting for something to change its screen.
rig provider switch --seat <seat> moves one seat to another subscription login on the same conversation, and rig provider precheck shows what it would do.rig provider switch --seat <seat> now moves one seat to another subscription login on the same conversation. A Codex seat moves to a login saved in its own Codex home (--to-account <name>, Linux only), after its current login is saved as before-switch-<time>. A Claude seat moves to its own subscription token file (--token-file <path>, made with claude setup-token), and --stored-login moves it back; OpenRig keeps only the file's path and your label, never the token. rig provider precheck --seat <seat> shows what a switch would do, changing nothing.
rig provider status --refresh can read a Claude token's usage on request.rig provider status --refresh --provider claude --claude-token-file <path> --credential-label <label> reads a Claude token's usage on request, one model turn per refresh.
A community-harness preset can set first_input: true to hand the harness its first startup turn at launch instead of pasting it into the pane. The Cursor example uses it and is no longer startup-assisted, and OpenClaw seats keep named conversations as a local agent or a Gateway client.
rig terminal open <view> --window opens the team's view there. A community contribution.Rex is a terminal provider. From a Rex pane, rig terminal open <view> --window and the TUI open the team's view as a Rex session; --provider rex selects it.
rig provider switch now acts; it used to answer failed_safely every time.rig provider switch acts: it used to answer failed_safely every time.
The Cursor reference preset is startup_assisted: false and passes its first prompt at launch.
An open rig view keeps a seat's tile when its tmux session is replaced, and a fresh replacement prints how to reopen the view.
The daemon starts listening before it works through a backlog of queue notifications, so startup doesn't time out behind them.
A startup database-conflict check that can't get an answer now says so, instead of reading like a clean one.
Proof-source monitoring watches the missions root, not the whole workspace.
Queue views, health reads with long histories and send checks do less repeated work.
XDG_RUNTIME_DIR used to let a seat's clients find the session bus; now the seat keeps DBUS_SESSION_BUS_ADDRESS=disabled:. To use a session bus, set its address in the rig's or member's env:. systemctl --user reaches the person's user service manager from a seat, whatever the bus address. A Codex stdio MCP server still needs env_vars = ["XDG_RUNTIME_DIR"] in Codex's config.toml to see the variable. (#2036)full_bypass). Handover and restore carry the rules, but aren't proven to keep the same conversation. An allow list doesn't withdraw broader inherited grants, and these are Claude's own tool rules, not OS or credential isolation. A change applies at the next launch or resume, not in a running process. Going back to code from before #1914 drops the stored rules from launch (see Before you upgrade). (#1914)rig seat restart <peer> --keep-conversation for each. A runtime switch starts a new native conversation (only the handover carries over), clears the source runtime's model, effort and Codex profile, and doesn't apply Claude launch additions such as an agent file or run startup slash commands. Selected separate-home restart/restore and useful conversation-continuity cases passed. Runtime switching worked in both directions in a minimal setup; one direction required an extra Enter. See Verification and remaining limits below for the untested paths.rig slack disable then enable for it, which marks waiting alerts as seen.rig daemon start can take 20 to 40 seconds before its listener warning, longer than the TUI's 30-second "restore everything" wait. On a slow host, run rig daemon start yourself first.claude setup-token) still gets no automatic usage reading. The reader accepts only claude.ai subscription logins and reports not_subscription_login for any other. Read a token's usage on request with rig provider status --refresh --provider claude --claude-token-file <path> --credential-label <label>; the label is yours, not an account the provider confirmed./usr/bin/env support for -0 and -u (option support checked on GNU coreutils and macOS 15; BusyBox and older macOS unverified). Without the required option support, launch stops instead of falling back to credential-bearing command-line arguments.⏸ manual mode on), an unframed box, and Codex seats are unchanged. Inspect the current native prompt and the refusal before retrying; do not assume that waiting 15 seconds clears it.lsof. Without a working lsof, the check can't tell whether another live daemon holds the database. The daemon now says so at startup and on /healthz, but the optional daemon.db_conflict: stop stops only on a confirmed conflict, so it doesn't stop then. The warning for a database outside OPENRIG_HOME doesn't depend on lsof. (#1974)policies.claude_compaction.message_file_path still asks once, and a packet read can ask after the native conversation changes. In 0.7.1's representative test, the restore also asked to run four read-only commands, which these grants don't cover. Check the retained restore map and reported gaps before treating restoration as complete. (#1940, #1952, #1962) A native check on Claude 2.1.309 passed the Read grants with no approval prompt; a complete restore with no help isn't claimed.rig compact <seat> --state and seat status now show the owed stage and its last delivery result; also read its saved restore map. Before sending a manual restore request, reconcile the native conversation and pending state to avoid a duplicate; do not repeat /compact merely because the restore request has not arrived. (#1953)haiku. That's Claude Code's own behaviour, and OpenRig now says so. Configuring the full id claude-haiku-5-5 gave a Haiku answer in a plan-mode follow-up on Claude Code 2.1.309. (#1942)rig queue create --host now reads INDETERMINATE, and the CLI reads the row back once by its ID. If it's still unknown, check that host for the original row ID before retrying. If the row exists, do not create a replacement. If a retry is needed, reuse the same --id and payload; an unsuccessful read is not proof that the row is absent. (#1911, #1978)$OPENRIG_HOME/plugins/openrig-core holds files that didn't ship, the Codex plugin still installs or upgrades, but OpenRig approves nothing itself: an existing approval stays as it was, and Codex's own hook review applies. On a restart without a version change, if a hook file or the plugin manifest there differs from what this version shipped, OpenRig doesn't install or upgrade its Codex plugin. On a version upgrade, OpenRig's managed refresh replaces an edited plugin file with the shipped one and keeps your edited copy under $OPENRIG_HOME/backups/managed-refresh/; after upgrading, inspect those backups and reapply any deliberate customization.needs-input), because OpenRig doesn't feed Herdr a state. Use OpenRig's view or the Feed to find waiting seats. Reopening a view with --provider herdr adds a workspace each time; from inside Herdr, rig terminal open <view> --window reuses a matching one. (#2032)Thank you to everyone whose work is in this release: @Vaishnavi220506, @Aschenbath, @dajiaohuang, @DeryFerd, @vanducng, @Nikhila-Dhevi-R, @Nikhi00718 and @Kero-66, and, for reports, ideas and outside audit, @borfast, @BillyOutlast, @tianqizhao0704, @calv-io-n, @george661, @fsafey, @astral-fleet-sessions and the Heretek-AI team.
Thanks also to @alexveli1 and @george661, who reported #1764 (addressed by #1778 and #1944); @kainne44 and @bobbiejaxn, who reported #81 (partly addressed by #1919); @oneandonlydean, who reported #1813 (partly addressed by #1951); @todie, who reported #1419 (its closure attribution addressed by #1811); @reisalbuquerque, who reported #49 (partly addressed by #1503); @jimallen, who reported #17 (partly addressed by #2028); @parkingmeter, who reported #1933 (addressed by #1963); @vanducng, who reported #1939 (addressed by #1964 and #1941, open for confirmation); and @haichongfu and @vadym-abstract-factory, who reported #14 (worked on again in #1921; still open).
Thanks also to @reco for the runtime-switch request in #589; @pr0xc3nt4ur1 for the edge-preservation and update case in #1193; and @m3ac-AllbrittenJ and @lev-gpt for the stall-reporting and age evidence in #331.
This release's selected checks found no demonstrated regression against published 0.7.1 or data loss. This is not an all-paths pass. The sealed package, installed checks and a real 0.7.1 upgrade passed at their recorded scope; both first-user journeys completed.
The long-paste issue #14 remains open; the Codex model-picker limitation #2010 and ordinary Codex resume/restore with a space in a workspace or queue-state path remain. The fixed in-place restart path is not proof for ordinary restore. Existing Linux session-bus, Claude member-rule, Pi recognition and other documented limits still apply.
Hermes, Cursor's plug-in work and the Codex member-permission counterpart are not included. Follow-up coverage review is scheduled for 2026-10-26 at 12:00 UTC; that is a review date, not a promise that every gap will be closed.
rig mcp serve --http serves one on another machine. A community contribution.OpenRig's MCP server grows from 23 to 34 tools, including identity, restore status, a rig's team and lead, following a handoff, reading a result and deciding inside a person's decision room. rig mcp serve --http serves an assistant on another machine, each caller acting as its own seat.
rig seat set-typing-guard. Input it holds stays readable and is never replayed.rig seat set-typing-guard <seat> --mode off|openrig-urgent|urgent-only|hold sets how much automatic input reaches a seat. rig send --urgent or an urgent queue row marks a message urgent, naming who marked it. Held input stays readable and is never replayed.
Asks to you. An ask lands on the TUI's Feed. Every ask still posts to Slack by default; rig gateway human set <id> slack-subscription urgent-only keeps Slack for asks marked --alert, which stand out with a mention.
rig skill loadout --seat <seat> shows them. A community contribution.Team skills live in each seat's own storage, not your project's .claude/skills or .agents/skills; rig skill loadout --seat <seat> shows them.
A daemon restart, such as an upgrade, picks up a managed compaction where it stopped.
Restart the daemon after installing. Eight database updates apply on its first start: run npm install -g @openrig/cli@0.7.1, then rig daemon stop && rig daemon start.
Startup replaces OpenRig's own plugin and skill files (openrig-core goes from 0.1.17 to 0.1.18). As in 0.7.0, it first copies each file it replaces under $OPENRIG_HOME/backups/managed-refresh/ and logs where, edited or not. If you did edit one, reapply your change from the copy.
Claude seats need Claude Code 2.1.280 or newer to see their team skills (claude --version).
acceptEdits, set it back once by hand. 0.7.1 stops changing it.If an earlier version reset your worktree's Claude permission mode, set it back once by hand. Launching a managed Claude seat used to rewrite defaultMode in .claude/settings.local.json to acceptEdits; 0.7.1 stops, but doesn't repair files already changed.
If you might go back to 0.7.0, undo Slack subscription choices first. 0.7.0 can't read a person file that holds one, so its people registry stops loading. Save each choice (rig gateway human show <id> --json > <id>.json), then run rig gateway human set <id> slack-subscription unset: someone on urgent-only gets every ask in Slack again, as 0.7.0 does. Set it again after you return to 0.7.1. The unset removes only the Slack subscription, so a person file that still holds a subscriptions: entry fails on 0.7.0 too. Don't restore an older gateway folder or database instead: that can lose newer people, bindings and delivery records.
If you go back to 0.7.0 and then upgrade again, old process evidence can keep a seat reading running instead of being relaunched.
rig seat set-terminal-policy. Existing seats keep full control.rig seat set-terminal-policy <seat> --preset no-peer-interventions|help-within-instructions|full-control sets how far OpenRig and other agents may act on a seat, with exceptions for compaction and model changes. Existing seats keep full control.
rig edge add and rig edge remove change one edge on a running rig.rig edge add and rig edge remove change one edge on a running rig.
rig attach as an outside app stays attached through rig down and restore.A seat joined with rig attach as an outside app stays attached through rig down and restore.
New seats get the daemon's OPENRIG_DB and TMUX_TMPDIR.
After a daemon restart that can't see the seats' tmux server, running agents are no longer detached and relaunched beside themselves.
A seat waiting at a prompt that only its pane shows, such as a Codex approval prompt, is shown as needing attention, with its reason, in the seat view, rig ps, the compact JSON and the rig count.
rig watchdog update changes a job in place, and the context watch takes token or percent thresholds. A community contribution.rig watchdog update retunes a job in place; the context watch takes token or percent thresholds.
rig stream subscribe and rig stream read, and rig stream show <id> says what became of one.rig stream subscribe <seat> and rig stream read give a seat bounded pages of the observations meant for it, and rig stream show <id> says what became of one.
rig project wake --decide, and gets back anything uncertain.Off by default: once decisions are enabled, rig project wake --decide lets your intake seat route stream observations with one bounded typed decision each, and hands anything uncertain back to that seat.
Stream retention is on: items emitted after the upgrade are deleted after 30 days, or beyond the newest 20,000, unless a record names them (retention.stream_days, retention.stream_max_items). History from before the upgrade is kept.
rig queue show --json drops lastNudgeInputBox, and send results no longer describe the input box.rig queue show --json drops lastNudgeInputBox, and send results no longer describe the input box.
An urgent or critical row's wake starts with Urgent (marked by …):.
rig queue show and rig queue whoami print notices for people on stderr.rig queue show and rig queue whoami print human notices on stderr.
rig send --dangerously-interact stops and shows you text already typed into a seat's input box instead of pasting over it.rig send --dangerously-interact --reason into a seat's input box stops and shows text already typed or pasted there, apart from suggestions, instead of pasting over it; clear it and retry, or pass --type-without-checking. Keys for a recognized native menu are sent as before.
rig send to a terminal seat types exactly your text, without the From and To lines, and queue wakes type nothing into it.rig send to a terminal seat types exactly your text, without the From/To lines, and queue wakes type nothing into it.
A stream body over 16 KiB is stored truncated and marked, and rig stream list --after from a pruned item returns stream_cursor_expired instead of an empty list.
A send held for a draft in the seat's input box reports inputPhase: "not_reached", and a managed compaction prompt held that way counts as not sent.
Reminders and workflow keepalives survive a seat swap.
One stuck row on a removed rig no longer stops the stuck sweep from flagging the rest, and stopped watchdog jobs release their old schedules.
A Claude member can bring its own agent file, system prompt file, plugin folders and Advisor model.
A Claude seat whose startup turn is refused reports its remaining startup context as owed; with an expired login, later startup inputs can still be submitted first (see Known issues).
A rebuilt Codex seat keeps its narrower sandbox, and read-only Codex profiles launch.
Upgrading OpenRig no longer stops a working Codex seat at "Hooks need review". OpenRig can renew approval of its own Codex plugin during an upgrade when the prior approval exactly matches the previous shipped bundle, nothing is disabled, and the installed bytes match the shipped bundle. Explicit disables and unrelated hooks are preserved.
An answer retried from Slack or Steering after an interrupted close is used once, and a replayed Slack reply stays absorbed.
rig acp run <qitemId> hands a task to an ACP agent such as Hermes or OpenClaw, and a new guide covers A2A agents.rig acp run <row> hands a row's task to an ACP agent such as Hermes or OpenClaw; a new guide covers A2A agents.
rig ps --nodes separates every column with a space and adds an ATTN column, so update scripts that read columns by position.rig ps --nodes separates every column with a space and adds an ATTN column before REASON; update scripts that read columns by position.
rig down asks before stopping the rig you're in, and without a terminal it stops nothing and prints the command with --yes.rig down asks before stopping the rig you're in; without a terminal it stops nothing and prints the --yes command.
rig up --plan no longer starts a stopped daemon; it exits with daemon_not_running.rig up --plan no longer starts a stopped daemon; it exits with daemon_not_running.
rig provider moves is back with a new job: filing and reading subscription move reports. It switches nothing.rig provider moves is back with a new job: filing and reading subscription move reports. It switches nothing.
From contributors: launch probe evidence (#1499, @Aschenbath); request IDs for creates and checked stops (#1790, #1792, @bradvin); faster cost history, ordered stream replay, kept web-UI drafts and guarded sends at Claude's MCP prompts (#1756, #1757, #1758, #1864, #1865, @Nikhi00718); a non-blocking preflight (#1759, @icey-zzz); clearer compact-plan output (#1629, @taljeri); exports keep culture files (#1657, @Totopo27); no flash in terminal views (#1663, @Vaishnavi220506); pod startup files for added seats (#1701, @AdamWhitehurst); Pi idle heartbeats (#1702, @okle100); the rig ps activity column (#1614, @shravansumanthanan); full workflow trails (#1732, @psbanka).
rig up <spec> --name <rig> runs one spec as several rigs. A community contribution.rig up <spec> --name <rig> runs one spec as several rigs.
rig bundle install <bundle> --plan --target <folder> previews the services an install would run. Proven on Linux.Off by default since 0.6.12: with bundle.services_enabled on (rig config set bundle.services_enabled true), a bundle carries its Compose file, rig bundle inspect shows each service before anything runs, and rig bundle install <bundle> --target <folder> starts its services, then its seats, in that folder. In 0.7.1, --plan --target previews the services an install would run, the before-install view also names a Claude member's own agent, prompt and plugin folders, and the example secrets-manager team publishes Vault's port on the host's loopback only. Proven on Linux; not run on macOS.
policies.claude_compaction.preparation_root puts restore maps where seats may write.
CLAUDE.md or AGENTS.md no longer overwrite each other's role, and rig down puts your guidance files back exactly as they were.A seat's role no longer overwrites the others' in a shared CLAUDE.md or AGENTS.md, and rig down restores your guidance files byte for byte.
Optional daemon.db_conflict: stop keeps a second daemon off a database another live daemon holds; the default still only warns.
rig seat status shows the permissions a seat's next launch will use and where each one comes from, and rig setup asks before adding to your shared guidance files.rig seat status shows the permissions a seat's next fresh launch will use and where each comes from; rig setup asks before adding OpenRig's line to your shared guidance files.
Community harnesses can report, name and reopen conversations, with examples for Cursor, Hermes and OpenClaw.
rig setup --providers is remembered, so OpenRig starts seats only for the tools you chose.rig setup --providers is remembered (provider.selected), so the kernel starts seats only for the harnesses you chose.
Herdr: tiles come back after a Herdr restart (one over SSH needs reopening), --session picks a Herdr session, and the TUI's terminal button works over SSH.
rig daemon start waits for a slow daemon instead of stopping it, and only warns when a secondary listener such as Tailscale's can't be checked. A community contribution.rig daemon start waits for a slow daemon instead of stopping it, and warns rather than fails when only a secondary listener, such as Tailscale's, can't be verified.
On Ubuntu 22.04's tmux 3.2a, a closing session no longer sends the query that crashes tmux.
OpenRig's hooks run without Node on the seat's PATH.
env command lines.Provider credentials and the activity-hook token stay off tmux, shell and env command lines; @pr0xc3nt4ur1 first raised it in #1360.
rig slack disable then enable for it, which marks waiting alerts as seen.rig daemon start can take 20 to 40 seconds before its listener warning, longer than the TUI's 30-second "restore everything" wait. On a slow host, run rig daemon start yourself first.claude setup-token) gets no usage reading. The reader accepts only claude.ai subscription logins and reports not_subscription_login for any other./usr/bin/env support for -0 and -u (option support checked on GNU coreutils and macOS 15; BusyBox and older macOS unverified). Without the required option support, launch stops instead of falling back to credential-bearing command-line arguments.rig acp run --help points to reference/acp-agents.md, which is not included in this release's reference context pack. Read daemon/docs/reference/acp-agents.md inside the installed @openrig/cli package instead.lsof. #1763's warning that another live daemon is using the same database, and the optional daemon.db_conflict: stop, find that daemon with lsof. On a host without a working lsof (each lsof call times out after 1 s), the check is skipped without a message, so neither the warning nor the stop happens. The warning for a database outside OPENRIG_HOME doesn't depend on lsof.openrig-core test on Claude Code 2.1.309 (acceptEdits), the restore and audit completed after three one-time Read approvals (the restore marker, post-compaction instructions and refocusing skill) plus four approvals for read-only commands. Check the retained restore map and reported gaps before treating restoration as complete.rig compact <seat> --state and its saved restore map. Before sending a manual restore request, reconcile the native conversation and pending state to avoid a duplicate; do not repeat /compact merely because the restore request has not arrived.reference/whats-new.md) overstates the hold behavior. Resolve the login and follow the seat's reported continuation guidance; this test did not verify recovery. (#1869)haiku and the plan permission answered on Sonnet in one observed run. Configuring the full id claude-haiku-5-5 gave a Haiku answer in the plan-mode follow-up.attention_required and needs one operator acknowledgement: rig seat clear-attention <seat> --reason "<why>". This wasn't rechecked natively on 0.7.1. (#1870)rig up --existing can then stop. Reported on 0.7.0; the same recognition checks remain in 0.7.1. (#1764)rig queue create --host request can time out or return failure even though the destination host created the row. Check that host for the original row ID before retrying. If the row exists, do not create a replacement. If a retry is needed, reuse the same --id and payload; an unsuccessful read is not proof that the row is absent.$OPENRIG_HOME/plugins/openrig-core holds files that didn't ship, the Codex plugin still installs or upgrades, but OpenRig approves nothing itself: an existing approval stays as it was, and Codex's own hook review applies. On a restart without a version change, if a hook file or the plugin manifest there differs from what this version shipped, OpenRig doesn't install or upgrade its Codex plugin. On a version upgrade, OpenRig's managed refresh replaces an edited plugin file with the shipped one and keeps your edited copy under $OPENRIG_HOME/backups/managed-refresh/; after upgrading, inspect those backups and reapply any deliberate customization.CODEX_HOME set in a seat's environment, rig seat restart --keep-conversation can report that the resume was not proven and that nothing restarted, although the process actually restarted and resumed. Inspect the seat's pane and native /status before retrying; a failed proof read is not proof that no restart occurred.rig provider switch does not switch a seat's account in 0.7.1. Its precheck can refuse an unknown target, and its execution path is not yet wired. The separate home-level Codex auth-switch/restart path is not this command.Thank you to everyone whose work is in this release: @Aschenbath, @bradvin, @Nikhi00718, @AdamWhitehurst, @puniserv, @pdurlej, @klueless-io, @shravansumanthanan, @taljeri, @Totopo27, @Vaishnavi220506, @okle100, @dajiaohuang, @icey-zzz, @akushonkamen, @psbanka and @adampog, and, for reports and ideas, @lev-gpt, @XVVH, @Kero-66, @pr0xc3nt4ur1, @VltrnOne, @BillyOutlast, @gogabrielordonez, @geoff-dhango, @alexveli1, @aanestesia, @team-sudarshana, @oneandonlydean, @t1m3fr33d0m, @smota and @z4cc.
Thanks also to @george661, who reported #1710 (partly addressed by #1721); @m3ac-AllbrittenJ, who reported #331 (partly addressed by #1801); @shintaii and @gopinathrimc, who reported #27 (partly addressed by #1735); @kainne44 and @bobbiejaxn, who reported #81 (partly addressed by #1876 and #1946); and @haichongfu and @vadym-abstract-factory, who reported #14 (partly addressed by #1871).
After a reboot, the daemon restores a kernel whose seats are all detached.
Quiet hours for the person. rig gateway human quiet set <id> --daily HH:MM-HH:MM (or --until) holds routine asks while the person is away and posts each one once when they're back. Escalations still go through.
rig archive, rig snapshot, rig export and rig restore --rig now take a rig's name instead of its id, and say so when two rigs share a name. A community contribution.Rig names everywhere. archive, unarchive, release, shrink, expand, snapshot, export, restore --rig and the matching MCP tools take a rig's name. A name two rigs share says so.
A seat missing a startup file no longer blocks the whole rig's restore; the other seats come back.
A busy daemon gets the real request; the CLI refuses only when it's confirmed down, and a timed-out write says its outcome is unknown.
Restart the daemon after installing. Six database updates apply on its first start: run npm install -g @openrig/cli@0.7.0, then rig daemon stop && rig daemon start.
If you pass accounts or API keys to new agents through tmux's global environment (tmux set-environment -g …), new agents no longer inherit it. Put the values in the seat's env, pass rig up --env KEY=VALUE, or keep the old behaviour with rig config set terminal.inherit_tmux_environment true. rig setup warns when it sees a custom tmux configuration. Running agents aren't affected.
Startup replaces OpenRig's own plugin and skill files with 0.7.0's copies. It first copies each file it replaces under $OPENRIG_HOME/backups/managed-refresh/ and logs where, whether or not you edited it: about 22 copies on the first start after upgrading from 0.6.14, so a copy doesn't mean you changed anything. If you did edit one, your change may stop working until you reapply it from the copy; the log line names the original file.
rig host doctor once for that host.If you write queue rows to another machine by its own id (member@rig@<self-id>), run rig host doctor <alias> once per host so the binding learns its address.
rig provider moves is gone, so a script that calls it needs changing.If a script uses rig provider moves, change it: the command and its setting are gone.
New agents start from a clean environment, not tmux's global one.
Seats added to a running rig get its culture and startup files (rigs created on 0.7.0).
rig up --env now reaches every way a seat can be launched.rig up --env reaches every launch shape.
Modes for the person. rig gateway human mode saves and applies a person's modes, ends them on a schedule with --until, and reports what each one changed. Inside a mode's agreed "decision room", the agent that asked can record a delegated decision; permission prompts are never decided this way.
rig queue show --json says why a wake didn't get through. A community contribution.rig queue show --json says why a wake didn't land.
A done close can't use --closure-reason blocked_on; use rig queue block <id> --on <child>.
Watchdog deliveries are recorded as they happened, and repeated refusals at a prompt raise an alert.
Before a new Claude agent's first startup message, OpenRig looks briefly (up to about five seconds) for Claude's input box, and keeps its checks on the submission. It doesn't guarantee delivery; see Known issues.
A new Claude agent's startup text can stay in its input box unsent. If a fresh Claude seat sits idle with text at its prompt, open its terminal and press Enter.
Codex seats that select openrig-core get its refocus and skills: OpenRig installs the plugin into Codex and, on a first install, trusts its own six hooks after checking their files, so a fresh Codex seat starts without Codex's hook review. The file check happens only when trust is granted: a later edit to the cached hook scripts keeps that trust, as with any hook trusted in Codex. The trust covers that Codex home, so your own Codex sessions there load the plugin too; its hooks do nothing outside an OpenRig seat. A plugin or hook you already disabled, or left untrusted, keeps your choice.
Slack: a thread reply answers only a question posted before it, and a rejected message isn't re-posted.
The TUI's agent table shows each seat's configured effort.
A long-open TUI says when its build differs from the daemon's, and refreshes less often.
rig provider moves and its setting are removed; report a subscription move as a task instead. rig slack verify drops --reason and --actor.rig provider moves and provider.subscription_moves are removed; report a subscription move as a queue item instead. rig slack verify lost --reason and --actor.
rig up --plan waits as long as a real launch, and rig seat restart and rig policy permissions apply now ask for permission.rig up --plan waits as long as an apply. rig seat restart, and policy permissions apply, now ask for permission.
rig gateway human list shows availability, rig attach prints Sender:, and a reused deliveryId no longer stops a send.Output shapes: rig gateway human list prints availability (available|focus|away|off); rig attach prints Sender:; a reused deliveryId no longer deduplicates a send.
rig ps --rig now filters correctly and says when it doesn't know a rig. A community contribution.rig ps --rig and --host --rig report an unknown rig and filter correctly.
rig compact --cancel stops a compaction that's already under way, without restarting the daemon.rig compact --cancel stops a managed compaction that already sent /compact, with no daemon restart.
A new agent's first startup message says it's OpenRig's identity context, not a task, and a seat with nothing else to start with is asked to confirm its identity and wait for an assignment.
Team agents are asked for permission less often: 15 housekeeping commands no longer prompt.
rig setup keeps 10,000 lines of tmux history, down from 50,000. A community contribution.rig setup keeps 10,000 rows of tmux history, down from 50,000.
env -u SSH_CONNECTION -u SSH_CLIENT -u SSH_TTY rig tui, or run rig terminal open '<view>' --provider herdr. A fix is on its way.rig daemon stop && rig daemon start). Don't use rig slack disable then enable for it: enabling marks any alerts waiting to post to Slack as already seen, so open exchanges miss them.tmux -V. A replacement version has not been verified against this crash.Thank you to everyone whose work is in this release: @Nikhi00718, @Aschenbath, @PROACTIVA-US, @jayhemnani9910, @rudycelekli, @markankamaransheogarat-sys, @haralez, @tianxingjian2406, @1solomonwakhungu, @AdamWhitehurst, @Quentin-M, @arxh80, @DeryFerd, @okle100, @zichen0116 and @13pgpg.
Tested on our own machines, and with first-time setups on Claude and on Codex, before release. A first-time setup on a Mac wasn't completed.
Agents with no recorded pane can be repaired. rig reconcile-session <seat> --repair-missing-pane --no-launch fills in only the missing pane after looking at the session twice. It keeps the seat's session, resume token and history, and adopts nothing. It refuses, changing nothing, when there's more than one pane or window, the pane changes between its two looks, a seat operation is running, or typing protection is on. If messages to an agent are refused with "Managed pane identity unavailable or changed; no input written." and that seat's pane was never recorded (seats an older OpenRig launched are often in this state), repair it in place, then send again; the repair needs typing protection off for that seat, so turn it off only if the person agrees, and back on afterwards.
The in-place repair refuses adopted seats and ambiguous panes. For a confirmed live seat, plain rig reconcile-session <seat> --no-launch re-adopts it into a new session row and attempts to capture its resume token; conversation continuity is not verified. If the intended pane is unclear, use rig context get help.
A getting-started page for people. docs/reference/quickstart.md explains how the pieces fit, the quick way in through the rigs skill, and what OpenRig changes on your machine. getting-started.md becomes the setup reference for the agent doing the setup.
rig context get reference/quickstart.md doesn't resolve. Read the page on GitHub or in the package's docs/reference/.
rig ask --wake refuses a runtime it can't wake (Pi, custom or unrecorded) instead of quietly trying a Claude wake. Upgrade the CLI and the daemon together to catch every case.
rig ps and the TUI read the list of agents through an index instead of scanning the whole event log.An indexed inventory read. The inventory read behind rig ps and each rig tui refresh now finds launch fences through an existing index instead of scanning the whole event log, and returns the same rows.
The refusals name the fix. The send refusal and the compaction refusal for a seat with no recorded pane now name the repair command, and the compaction refusal explains how to turn typing protection off and on with the person's agreement. rig config --help lists the new Mac setting, rig ask --help explains its exit codes, and the CLI reference and the openrig-user skill list the repair.
The bundled openrig-core plugin goes from 0.1.14 to 0.1.15, and /healthz gains an euid field beside pid.
Refocus does less work on a busy machine. It finds each notes file itself instead of starting the CLI once per level.
A manual compaction waits through an approval. When a seat stops at an ordinary approval prompt while preparing, rig compact keeps waiting inside its deadline instead of giving up. OpenRig never answers the approval itself. A manual rig compact can take longer to answer: its preparation gets 180 seconds (was 120) and the request 240 seconds (was 180).
After a compaction, confirm it ran by the restore request or the drop in context, and check that refocus showed the agent's current work.
An optional Mac check, off by default. On a Mac shared by several accounts, rig config set transport.local_user_guard true makes the CLI refuse a local daemon that another account runs. It's a guard against reaching the wrong daemon by accident, not authentication.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. No database migration runs. Coming from 0.6.12 or earlier, read 0.6.13's notes first: its migration runs on the first start.
The roadmap drops three finished items.
rig reconcile-session <seat> --no-launch re-adopts it into a new session row and attempts to capture its resume token; conversation continuity is not verified. If the intended pane is unclear, use rig context get help.rig commands can reach the first user's daemon on the default port unless the Mac check above is on. Running each person's daemon on its own port lets both work, but doesn't keep them apart.rig up <rig> --existing can refuse to restart a rig after one of its seats was stopped on purpose. Use rig restore <snapshot> --rig <rig> instead; the stopped seat stays awaiting your decision.rig tui keeps the daemon busy, re-reading every 30 seconds, and keeps its old code after an upgrade. Quit it when you're done, and reopen it after upgrading.rig context get reference/quickstart.md doesn't resolve. Read the page on GitHub or in the package's docs/reference/.See which model a seat was actually using. rig ps --nodes, rig roster show and rig roster find show the model each seat's current agent was last seen using, beside its configured model, with where and when it was read. Claude and Pi readings come from the agent's own responses; Codex's come from its recorded turn settings. Missing or stale evidence reads unknown, never the configured model. Plain rig ps --nodes prints one model line per seat after the table, and JSON gains modelObservation.
Move a seat to a new working folder: rig seat set-cwd <seat> --cwd <absolute-path> --reason <text> saves an existing seat's working folder for its next fresh launch or rebuild handover, and keeps its address, queue and history. It refuses when files already in the new folder would clash with the startup files the next launch writes. The running agent doesn't move: switch with rig seat launch <seat> --fresh --stop --reason "<why>" or a rebuild handover. rig seat status shows the saved folder.
rig seat set-cwd checks the new folder once: a refusal can leave the folder created, and background work, such as a Claude seat's collector files or a managed compaction's restore map, can use the saved folder before you switch. Switch promptly, and don't run rig down/rig up or a fork handover first.
A Claude seat handed over in place waits for the new Claude. A fresh handover that reuses the old occupant's pane could read the old Claude's screen as the successor being ready, and type the handover packet into a shell before the new Claude had started. A ready result taken from that history is now checked against the current screen, so the packet waits. This corrects the common case.
A fresh in-place handover of a Claude seat can still report ready early in some cases: a screen without a header whose process list can't be read, a Claude process seen but not proven, or a pane or screen that can't be read. Check that the successor read its packet.
Slack recovers what it missed in more places. Missed-message recovery used to read only the configured channel's top-level history, so a reply in an older thread or a message in a mapped channel was lost for good. It now also recovers the channel map's channels from when each was adopted, replies in tracked threads whose root was posted or which closed in the last seven days, and, on a machine without its own Socket Mode connection, each channel the hub has relayed to it, from that channel's first relayed event. A recovered reply lands as a late row in its thread and never answers a decision; the seat reads it and decides. A Slack rate limit now holds that method for every channel, or every thread, and is kept across a restart. rig slack status shows a line per channel and a thread line. On its first start, threads inside the seven-day window are scanned once from their root; replies that already landed are recognised and not delivered twice.
Missed replies in older open threads aren't caught up; slack-thread-coverage.json isn't pruned yet; a default-channel backlog can drain at up to about half its earlier rate while threads also need calls; and a downgrade to an older release and back inside a rate-limit window can let other channels call Slack during that window.
rig grow <rig> and rig remove <rig> <node> accept a rig name as well as its id; a name two active rigs share is refused with the candidate ids.
After a failed restore, a fresh launch that reports ready clears the old restore warning: rig ps and rig seat status stop showing attention_required from it, and rig seat clear-attention no longer refuses.
A Claude Code or Codex seat showing a complete idle screen reads idle in JSON and filters even when the terminal repaints, instead of running.
Parked rows for one seat whose wake timers fall due in the same pass now arrive as one message naming every row, instead of one interruption each. A single reminder, a custom --wake-message and repeating waits keep their own message.
A rig send --wait-for-idle whose caller went away (Ctrl-C, or its own timeout) used to keep waiting in the daemon and type once the seat went idle; if no input has started, it now types nothing, then or later, and is refused with caller_cancelled. The same holds for broadcast recipients not yet reached and for Enter-only sends.
When an agent closes an item addressed to a person and the follow-up can't be verified, the stuck-item finding goes to the agent that closed it instead of to the person.
rig queue list --state "pending, in-progress" with a space after the comma now matches both states, as does the MCP rig_queue_list tool.
rig restore-check can report caveats where it used to be green. Its hook check now tells apart a saved hooks resource, the core plugin's relay, and neither, and reads a broken or unknown plugin selection as a caveat. It still exits 0.
Managed Claude seats get your editor and display. Seats launched with a native permission mode now receive EDITOR, VISUAL, DISPLAY, XAUTHORITY and an absolute WAYLAND_DISPLAY, so Ctrl+G opens your editor and image paste can work. A running seat picks it up at its next launch (rig seat launch <seat> --fresh --stop --reason "<why>", or rig down then rig up --existing). A git commit without -m inside such a seat now opens your editor.
The refocus work packet stays off for both Claude Code and Codex. With OPENRIG_REFOCUS_WORK_PACKET=1, each field is now bounded: one-line summaries of about 160 characters with a rig queue show <id> pointer, at most 20 held and 10 next rows with a count of how many are shown, and section bodies cut at about 1,500 characters. There's no total size cap.
The bundled openrig-core plugin goes from 0.1.13 to 0.1.14. The Claude activity-relay and context-collector hooks OpenRig writes into a project's .claude/settings.local.json now call the daemon's own Node by its absolute path instead of bare node, so a Claude session with an older PATH stops showing "node: command not found" after each reply. Existing bare-node entries are refreshed automatically, and your own wrappers are kept.
The plugin's own manifest hooks still call bare node, so a Claude session with an older PATH can still show a Stop-hook error from them, and two plugin hook runs for the same seat aren't serialized.
Claude Code 2.1.303's MCP-server approval menu now reads as needing input, so a restore doesn't call that seat ready and an ordinary send no longer pastes into the menu.
A fresh running hook still sends into Claude's MCP approval menu.
A Claude launched through a wrapper script (a claude on PATH that execs the real binary) is now proven by its executable path, so a restored auto-mode seat whose banner has scrolled away resumes ready instead of needing attention.
When rig seat restart --keep-conversation stops at a Codex startup screen, the report names it (hook review, folder trust, model choice or update), says who decides, and points to rig capture and the recovery command.
rig seat restart --keep-conversation --help says an adopted seat refuses with claimed_session and which conversation a restart can prove.
Codex's current "Trust this folder?" panel reads as a folder-trust gate instead of booting or ready, so startup text and Enter don't reach it.
Pi seats now use their effort. An effort already in a spec reaches Pi as --thinking <level> from each seat's next managed launch, restore or handover, and on every managed launch and resume after that. rig seat set-effort and rig seat effort now work for Pi: the report shows what was requested, saved, sent and, on Pi 0.99.0 or later, recorded.
An OMP seat whose turn ended with a typed transient provider failure (status 408, 429, 500, 502, 503, 504 or 529) can take an ordinary rig send retry instead of staying in needs-input; the error stays visible in its pane, and nothing retries on its own.
An OMP seat releases needs-input for any failure carrying one of the seven transient statuses, including an authentication failure that carries one; untyped errors and Pi seats keep attention.
A Pi or OMP seat's runner prints its READY line once per ready transition.
The TUI's Feed shows open updates addressed to a person as "No action needed" cards, critical first, even without a Slack delivery receipt, and no longer shows one twice.
rig capture notes on stderr whether the text in a Claude Code or Codex input box looks typed or looks like a faint harness suggestion; stdout is unchanged, and --json and the MCP tool carry an inputBox field.
Each account row in rig provider status, its JSON and the TUI usage view shows a sign-in fact (usable, needs sign-in or unknown) from evidence a refresh already holds; a signed-in Claude login never reads better than unknown. A signed-out Codex login reports not_signed_in instead of rpc_error.
With provider.usage_windows on, a Codex account's last reading, its reset and its last observed exhaustion survive a daemon restart and a login switch. After a restart, rig provider status shows the last reading as history ("kept from before the daemon restarted; not a current reading") with availability unknown until a fresh read; when a window's reset has passed, it shows an expectation that the window is full again, which is planning, never a reading. Rows add accountHistory, lastObservedExhausted, retained and savedProfile.
The capability map explains comparing token use across machines with rig usage top.
rig doctor's cmux warning now gives a fix that works.rig doctor's cmux warnings give a fix that works: set Socket Control Mode to "Automation mode" in cmux, then run rig doctor again.
rig provider prints the daemon's error message instead of a bare code, and rig compact's refusal for an unrecorded protected pane names rig reconcile-session <session> --no-launch. The README's "What OpenRig changes on your machine" says when Codex hook configuration is written, that your own Claude status line is kept, and which settings files are preserved.
A rig name with a dot is refused before anything launches. tmux turns . into _ in a session name, so a seat in demo.v2 ran as …@demo_v2, where OpenRig couldn't reach it, and its tmux session was left behind. rig up, adding a member, a first-start retry, bundle install, expand and the preflight preview now refuse with a message naming the name tmux would create. An existing rig whose name has a dot gets this refusal when you add a member or retry its first start.
The node-detail API's targetPath names the file a guidance block is really in, and only while that file holds the block, instead of echoing the spec's declared target.
The web UI shows an aborted workflow as ABORTED, not COMPLETED.
A redriven workflow no longer keeps its old completion time while it runs again.
The warning for an edited projected file gives remedies that exist: fold the edit into the spec, or move the file aside before the next projection. There's no --force flag.
The shipped agent specs mark their role file as the seat's role. After a seat's next fresh launch, rig queue whoami --json reports its role instead of not-declared, and refocus names the role file. The agent startup guide shows how to remove a leftover shared role block that a kernel installed before 0.6.5 may have left in CLAUDE.md or AGENTS.md.
Getting started lists Rig HQ, a community-built read-only viewer for rigs; OpenRig hasn't tested it.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. One database migration runs on every upgrade: 117 adds an empty, Codex-only provider_account_readings table. It fills only with provider.usage_windows on (off by default), and then keeps each resolved Codex account's usage readings across a restart. There's no migration 116. If you're coming from 0.6.11 or earlier, read 0.6.12's notes first: its twelve migrations run on the first start.
Lighter on Linux: the daemon reads /proc itself, on a worker thread, instead of forking ps to list processes. The author measured CPU per listing dropping from about 69 ms to 8.4 ms, about 1.5 s of CPU a minute less at rest. macOS still uses ps.
The one-line installer no longer calls an unfinished setup a failed install. When rig setup (step 4) doesn't finish ready, it now ends with INSTALLED, SETUP NOT COMPLETE (exit N)., the CLI's path, whether the Node.js and SQLite check passed, a pointer to setup's "Some steps need attention" list, and "run rig setup again; the CLI does not need reinstalling". Setup's exit status is kept, and steps 1 to 3 keep their FAILED [n/4] line. No dependency changes.
Getting started and the README: one working Claude Code or Codex account is enough, and the rigs skill lists what starting OpenRig writes.
On a Mac where you aren't an administrator, or where Homebrew belongs to another account, the docs say to install Node.js with nvm in your own home. The publishing guide says what the registry's checks decide: a submission whose name is taken is still received, with a hint; a team sharing a listed team's rig.yaml name passes with a notice for the maintainer; and listing is the maintainer's decision.
Regression checks tie a managed /compact to its hook effect and parked work to the refocus context; a test fixture no longer expires on a calendar date; and development main carries 0.6.12's versions and release text.
rig commands can reach the first user's OpenRig. A second user's commands can reach the first user's daemon on its default port (7433). Running each user's daemon on its own port lets both people work, but it doesn't keep them apart. This release doesn't change it.rig send with "Managed pane identity unavailable or changed; no input written." Nothing is typed into the seat. Seats that 0.5.x launched recorded their tmux session but not its pane, and 0.6.x needs the pane before it sends. This was seen upgrading one install from 0.5.0 to 0.6.12; that install stays rolled back, and older installs weren't qualified for this release. Not every older seat is affected: one that was adopted or rebound since can already hold its pane. The same code is in 0.6.12, so it isn't new in this release, and this release doesn't include a repair. Don't upgrade an install whose seats a 0.5.x or older OpenRig launched without checking that they still take rig send.partially_restored: on 0.6.12 and again on this release's tests, the seats meant to be running came back (three with their original conversations here), and a deliberately stopped seat was left awaiting a decision. Launching that seat fresh works, but it starts a new conversation.node, so a Claude session with an older PATH can still show a Stop-hook error from them (#1344, #1475, #1545).slack-thread-coverage.json isn't pruned yet; a default-channel backlog can drain at up to about half its earlier rate while threads also need calls; and a downgrade to an older release and back inside a rate-limit window can let other channels call Slack during that window (#1536).rig slack verify on the hub doesn't check relayed channels, and an answer from a seat other than the one the message reached still posts at the top level (#1436).rig seat set-cwd checks the new folder once: a refusal can leave the folder created, and background work, such as a Claude seat's collector files or a managed compaction's restore map, can use the saved folder before you switch. Switch promptly, and don't run rig down/rig up or a fork handover first (#1498).rig policy current and plain rig ps don't show a policy stored with rig policy apply --seat; rig seat status and pod-aware export do (#1285).env: and --env values aren't secret, members added later with rig grow can't carry a member-level env:, and rig up --env is not forwarded for flat-node specs or v1 bundles (#1294).unknown with usage windows on, and its sign-in fact never reads better than unknown (#1504, #1508).openrig-user skill's subscription section mentions apps that don't ship; don't look for them (#1450).bundle.services_enabled on, rig up <file>.rigbundle installs into the current folder when you don't give --target (#1474).rig seat restart --keep-conversation supports Linux only.blocked because of one seat that isn't running (#1141).rig scope slice create says it couldn't verify membership, the slice was kept. Check mission.yaml before creating it again (#1007).rig doctor checks both harnesses; rig tui --shared attaches on the default tmux server; a team's first steps can still ask for approval; a seat waiting at a permission prompt shows no product notice of its own; a launched team writes into its working folder; a resumed Claude seat can't be confirmed after /clear, an in-process /resume or a hand-started process; Codex team seats don't ask before lifecycle commands; the send gate trusts the latest recorded hook; and after a restart rig status shows Kernel: skipped.claude alias defined only in a nushell rc doesn't resolve; after a reboot, start the daemon with rig daemon start and bring your rig back with rig up <name> --existing; the web UI's file preview runs scripts, so preview only files you trust; browser access isn't complete browser isolation; and Codex network inside the sandbox hasn't been checked with real logins, managed policy bundles or on Linux.An agent with no shell can hand work to the team through rig mcp serve, which gains five queue tools (rig_queue_create, rig_queue_show, rig_queue_list, rig_queue_claim and rig_queue_update), acting as the seat named by OPENRIG_SESSION_NAME. Nothing wakes such an agent; it sees new work when it checks.
rig usage cost [--seat <session>] [--json] shows what each seat's harness reports about cost beside its tokens: for Claude, the status line's own estimate for the current session (labelled an estimate, not an invoice); for Codex, tokens and a note that Codex reports no cost. A missing figure reads "not reported", never $0. It needs no setting, and the TUI's new System → Cost page shows the same lines.
Claude seats now run the hooks of the bundled openrig-core plugin, which goes from 0.1.9 to 0.1.13. A Claude seat whose plugins include openrig-core (every agent spec OpenRig ships selects it) starts with CLAUDE_CODE_PLUGIN_DIRS pointing at OpenRig's copy, so its activity relay, compaction save and restore hooks, and refocus actually run. Before, the plugin was copied to a folder Claude Code doesn't load. It needs Claude Code 2.1.280 or later; an older Claude Code ignores the setting, and OpenRig doesn't check the version. A running seat picks it up at its next launch, restore or handover. The refocus hook may now take up to 60 seconds instead of 5, so a refocus that hangs can hold up a prompt for up to a minute.
Two plugin hook runs for the same seat aren't serialized: if a harness overlaps them, a post-compaction record can be lost or a refocus delivered twice. Overlapping runs weren't tested, and the samples that showed no prompt stall don't show that overlaps or stalls can't happen.
rig start --last leaves down a rig you stopped on purpose with rig down or rig seat stop, and prints how to bring it back (rig start --rigs <rig>); --json carries keptDown. --all, --rigs and the picker are unchanged. After rig down, rig ps and the TUI read the rig stopped instead of recoverable, and the printed restore command is rig up '<name>' --existing, which works even when the name matches a shipped starter.
Set one seat's reasoning effort: rig seat set-effort <seat> --effort <level> --reason "<why>" (or --clear) saves it, and it's used at the seat's next launch, restore, handover or in-place Codex restart; nothing restarts. rig seat effort <seat> reports what was requested, what's saved, what the running process was launched with and what the session recorded.
Change an existing seat's permission policy: rig policy apply <policy> --seat <seat> --reason "<why>" stores it for the seat's next restore, resume or launch. The running agent and the spec file don't change.
rig policy current and plain rig ps don't show a policy stored with rig policy apply --seat; rig seat status and pod-aware export do.
rig launch takes a rig's name as well as its ID, locally and with --host. A name two active rigs share stops before anything is sent and lists their IDs (rig_ambiguous). Before, it took only an ID, and a name failed with not found.
An ordinary terminal seat relaunches fresh on restore and re-runs its startup actions, instead of stopping at awaiting-decision on every restore after its first. A required startup file missing for such a seat now blocks the whole restore before anything launches (required_startup_file_missing).
Restore stops a seat whose current runtime differs from the one its snapshot started with, before creating its terminal session, and --fresh doesn't bypass it.
A slow start ends as needing attention, not failed. A restored Claude seat that hasn't drawn its screen is watched for up to runtime.readiness_timeout_seconds (30 s by default) and then ends attention_required with its pane and attempted resume token kept, so a restore can take longer: up to about 27 seconds more for each seat that never starts, one after another. A seat is described as "returned to a shell" only when the process tree shows an idle shell, and the older restore path for rigs without pods keeps such a seat for attention instead of rolling it back.
A restore starts seats one after another, so each seat that never draws its screen adds about the readiness timeout.
rig add, single-seat rig launch (including --retry-startup-from) and rig launch --seats wait up to 120 seconds for the daemon instead of 5. A timeout there still isn't a rollback, so check rig ps before running it again.
rig remove and rig shrink keep a seat that receives work while it's being stopped: the late work goes to your --fallback, or the stopped seat is kept and the command answers 409 with the work listed. A stale rig unclaim that would detach the seat's newer occupant is refused with 409 session_not_current.
A seat whose startup failed for a recoverable readiness reason becomes ready after a successful send or capture.
A periodic snapshot taken while rig down was stopping a rig no longer replaces the complete pre-down snapshot.
A failed process-lineage check says what it found (for example, that the running Codex was launched with a different conversation than the seat records), instead of blaming the pane's foreground label.
rig reconcile-session no longer reports a runtime drift for a pane reading sh when Claude or Codex is running in it.
rig seat set-resume-token writes to the seat's latest session even after the system clock moved backwards.
When rig seat clean refuses because an older session is still alive, its advice names that session instead of suggesting rig seat stop, which would act on the current occupant.
rig queue update <id> --summary "<text>" on its own corrects a row's subject in any state.
rig queue update <id> --state <closing state> --resolution "<outcome>" stores the outcome on the row, rig queue show returns it, and a later fallback adds its line instead of erasing it.
rig queue show <id> --full on a closed row adds closingTransition: its state, reason, target, who closed it, when, and the closing note.
Off by default: transport.verify_input_box (OPENRIG_TRANSPORT_VERIFY_INPUT_BOX). A --verify send to Claude Code or Codex reads verified only when the text was seen in the input box and the box was empty after Enter. It only reports; nothing is retried. Multi-line sends and wakes to Claude Code stay unconfirmed.
Per seat, off by default: rig seat set-typing-guard <seat> --mode draft-aware --reason "<why>" checks the input box before typing and before Enter, and holds or refuses a message instead of typing over someone's unsent text; nothing retries it. --mode hold is the existing on, and --mode off the default. The draft-aware guard's shared send and startup code runs for every seat.
The draft-aware typing guard waits for idle inside the seat's delivery lease when the input box is empty, which can delay other work for that seat (not measured), and it can't guard text a person types in the instant before OpenRig types.
output.compact (OPENRIG_OUTPUT_COMPACT, off by default, from 0.6.11) now also shortens rig queue show and rig queue transitions (--full prints everything; rig queue transitions --full is new), and bounds rig whoami --json peers to output.whoami_peers (new, default 8), connected peers first; --all-peers lists everyone. Over SSH --host, the note's full-list command now names the host.
rig send --dangerously-interact types exactly the bytes and keys you give it, in any state and over every transport: a lone CR, Tab, Space or Esc acts as that key. Enter is added only when a complete answer is visibly staged in a recognised Claude Code or Codex input box; anywhere else, put CR in the payload yourself.
rig queue update refusals you can fix answer HTTP 400 and exit 1, not 500 and exit 2: eleven error codes for closure fields, blocked_on or wake options where they can't be stored, and a watchdog that isn't live or targets someone else. blocker_not_found, blocker_not_live and blocker_malformed still exit 2.
rig seat status reads Typing guard: on (hold) for hold, and its JSON's desired and effective booleans now mean hold only.
Closing a row as handed off without a verifiable successor now points to rig queue handoff, and the stuck sweep flags a row closed onto itself.
Help, warnings and the shipped skills say --summary is a row's subject everywhere it's listed, and carry the project:, mission: and slice: tags; the reply hint for messages from a person includes --summary. The TUI's RECENT list shows the first line of a row's body when it has no summary.
With nobody registered, a send to a person says to register one with rig gateway human add instead of reporting a missing registry projection.
The coordination notes say which queue fields answer pane delivery.
A send to a pane in tmux copy or view mode leaves the mode once and delivers. A pane that can't leave its mode gets no text, and the send is refused with pane_in_mode (HTTP 409). A person scrolled back in that pane is returned to the bottom.
A held-back wake no longer uses up a one-shot park timer (0.6.11's known issue): a wake the typing guard holds back is tried again at the timer's interval, and a shared one-shot timer retires when its last attachment is consumed.
Stopping or retiring a watchdog while a check is running stops it: no late wake and no late record; its last attempt's time is still recorded.
A delivery already in transport when its watchdog stops still completes, with its records.
A waiting row whose blocker changes while an earlier wake is being sent still gets its next wake.
A row parked on its owner's own repeating reminder isn't counted as stranded once the reminder has delivered, and rig parked marks it servedByOwnReminder.
A cross-host handoff's closing record carries the sender's identity, as a local one does.
Absorbing an inbox entry into queue work is one transaction, so overlapping absorbs make one row.
A seat can give a slow refocus more time with OPENRIG_REFOCUS_BUDGET_MS, from 4500 (the default) to 52000 ms.
A new page explains how to keep your own Claude status line with OpenRig's context collection.
The fable model alias now names claude-fable-5-1, so a fable pin on a seat running claude-fable-5 raises a divergence notice. A Claude pin ending [1m] no longer does when the seat runs that model. Pinned Pi seats are now checked, and a Pi thinking-level suffix such as :high doesn't count as a divergence.
With managed compaction on, Claude sessions OpenRig launches carry idleCompaction: false, so Claude Code 2.1.290 or later doesn't compact them on its own while idle.
With usage windows on, a Claude account's current availability always reads unknown, because nothing links a seat's reading to the login it used; each window still shows as that seat's reading.
Managed /compact is kept to 600 characters, sent only to an idle seat, and reported separately. Since 0.6.10 the managed command had grown long enough that Claude Code took it as pasted text, so it didn't run while the attempt read as delivered. The rest of the instructions move into the restore map and the restore request. rig compact and rig compact --state show a separate compact result: submitted, unconfirmed (it reached the seat; that isn't proof compaction ran), uncertain (not replayed) or not sent. A running draft-aware seat with no recorded pane is refused with pane_unknown_protected: rig compact exits 1 (was 2) and the route answers 409 (was 502). Over the shared transport, a write that throws without a code makes rig send exit 1 with HTTP 409 instead of 2 with 500.
The command budget is 600 characters, set from one comparison on Claude Code 2.1.282 (792 characters ran as a command, 850 didn't), and a later Claude Code could move that line. Nothing confirms compaction ran; read submitted, unconfirmed as reached, and confirm by the restore request or the drop in context.
Claude login detection is narrower. Only a current refusal (Not logged in · Please run /login, Login expired · Please run /login or OAuth token revoked · Please run /login) reads login_required. The "Not logged in" footer alone no longer does, so a seat signed in through a gateway starts, resumes and restores normally.
Automatic compaction ignores a context reading taken before a seat was stopped, replaced, handed over or restarted.
When a Claude seat's startup prompt is still unsent, the launch presses Enter up to twice, each time only while the box holds exactly that prompt, and more Claude footers are recognised, so fewer healthy launches warn "Startup submission unverified".
In a Claude seat started with a permission mode, Shift+Enter inserts a newline when tmux's extended-keys is on: the launch passes the pane's TERM_PROGRAM and TERM_PROGRAM_VERSION to Claude. The getting-started guide says how to add a newline when Shift+Enter submits.
A launch leaves a project's .claude/settings.local.json or .mcp.json untouched, with a warning, when it isn't a JSON object, instead of overwriting it.
OpenRig no longer rewrites a Claude seat's settings file every 30 seconds when its status line is already in place, and replaces a read-only copy of its context collector instead of failing.
Off by default: provider.subscription_moves (OPENRIG_PROVIDER_SUBSCRIPTION_MOVES). A record of moves of a Codex home's seats between subscriptions, rig provider moves [--home|--seat|--limit], rig provider moves show <id> and rig provider moves record --file <json|->, plus a section of the openrig-user skill for an agent a person asks to manage subscriptions. While it's off, the commands exit 1 and the routes answer 409.
The openrig-user skill's subscription section mentions apps that don't ship in this release; don't look for them.
The agent startup guide explains how a Codex seat compacts and how to size it.
The daemon writes Codex's config.toml only when a Codex seat exists or launches, so a daemon that never runs Codex leaves ~/.codex/config.toml alone.
The daemon logs why it couldn't capture a Codex seat's resume token, once per reason, and stops those captures when it shuts down.
A Codex seat that forks a conversation keeps its startup context when the new conversation's ID arrives late. The launch waits up to the readiness timeout for a verified child; if none arrives, the seat stops at attention_required, and rig seat continue <seat> delivers its role once. A handover to a forked Codex conversation completes with the ID marked pending instead of failing.
A Codex seat started through a shell wrapper isn't reported as returned to a shell while it starts.
The Codex resume command OpenRig prints keeps --no-daemon when the seat's last managed launch used it; for seats launched before this, the guidance says the choice wasn't recorded.
Codex's » prompt is recognised when reading an input box, so a staged message on a Codex seat can get its one guarded Enter, and rig send --verify tells a message still in a Codex input box from one that was delivered.
A Codex seat reads as needing a person only when its pane shows a real approval or selection prompt; an unreadable pane still gets there after a short grace, as permission_request_unverified.
Off by default: usage.pi_cost_facts and usage.pi_openrouter_charge (OPENRIG_USAGE_PI_COST_FACTS, OPENRIG_USAGE_PI_OPENROUTER_CHARGE). Pi's own cost estimate in rig usage cost, counted once with a fork's inherited entries kept apart, and OpenRouter's actual charge beside it, never added to it. With both on, the Pi runner asks OpenRouter for each request's charge with the key OpenRig already gives the seat. A Pi seat reads both switches when it launches or resumes, so a change takes effect then; what was recorded stays visible. With every setting off, a Pi models.json notice can report a provider key's presence instead of unknown, because models.json is now read the way Pi reads it.
A Pi or OMP seat whose agent has exited to a bare shell reads attention_required (process_identity_mismatch) instead of running, so rig start offers its rig as a restore candidate.
The rig spec reference says a HOME value reaches Pi's agent.
An OMP seat's restore catch-up and history refresh work past 1 MiB of history when OMP offers its RPC protocol v2, up to 64 MiB.
A failed Pi or OMP turn shows the first usable error field.
One Slack app can serve several machines: the machine holding the app-level token relays the channels listed in forwardChannels to the machines that own them, and a person's answer posts in their own thread. A person's new message in a rig's own mapped channel now goes to that rig's representative (or a seat named with --inbound), and every seat post's sender line names the machine (from <seat> on <host id>), on one machine too.
One Slack app across machines was simulated in one process, never run on two real machines. Relayed events are trusted from the calling daemon, a relay that keeps being refused is retried with no limit, rig slack verify on the hub doesn't check relayed channels, and an answer from a seat other than the one the message reached still posts at the top level.
A botInbox block delivers named bots' posts in one Slack channel to one seat, as rows marked as data from a program. The app manifest asks for one more scope, users:read.
rig slack manifest --help and the setup guide say how to paste the manifest when Slack shows it empty.
A workflow gate waiting on a person shows that person the question, not the agent packet, and still posts when its tags are very long.
An ntfy notification title with an em dash, an emoji or other non-ASCII text is delivered as written, and shortened notification bodies, slice stories and rig ask excerpts keep emoji whole.
The TUI's slice header says what its proof count counts: accepted with a judgment policy, paired without.
TUI helpers keep working after the folder you started it in is deleted.
In the TUI, Open terminals sends the window's width, so a wide terminal opens the default view; the file reader finds headings with combining marks; the Help list stops at its last command; a sideways trackpad swipe no longer scrolls up or down; and rows cut to width keep emoji whole.
GET /api/events takes ?replay=0|<n>|all and streams its replay 500 events at a time; the default is unchanged.
provider.usage_windows (OPENRIG_PROVIDER_USAGE_WINDOWS, daemon, off by default, from 0.6.11) now also shows Claude's 5-hour and weekly windows from the seats' status lines in rig provider status, with --provider codex|claude to limit a refresh; each seat's configured account; and, with --all-hosts or --hosts <ids>, every registered machine's windows, listed once per allowance and never added up. The TUI's System → Usage and Usage · all hosts pages are always listed and, with the setting off, say how to turn it on.
rig ask exits 1 for an unknown or ambiguous rig instead of 0.
The CLI starts faster: it loads only the command being run and turns on Node's compile cache, so rig queue list and rig ps start several times faster in the author's measurements.
rig create --json says why a create failed, in detail.
rig status --json prints the daemon, kernel, workspace root, rigs and cmux state as one object; daemon.reachable: null means unconfirmed, not down.
rig capture --lines N that cut lines says activity is unknown, on stderr and as activity: "unknown" in --json. rig ps with a context-usage filter but no --nodes says on stderr that the filter wasn't applied.
rig mode keeps a saved mode and a work posture side by side at one target, and rig mode unset --category mode|posture removes one; the rig mode set preview reads an unusable answer as unknown; and rig mode's human output marks the eight saved fields nothing in OpenRig acts on yet.
rig usage top no longer counts a new conversation, or a reset window, as burn. Tokens and rates are measured within one history and one window, so figures can drop and rankings change, and it no longer shows a negative rate after a window resets.
A stopped or removed seat's transcript capture stops, and its saved transcript stays readable at any age. A quiet seat's transcript health can read live with reason capture_unchanged while its last capture is more than 10 seconds old. rig transcript --tail N no longer cuts off the first requested line.
GET /api/hosts reads an IPv6 HTTP host as reachable when it is. The web UI's Log page keeps emoji whole in shortened previews.
rig usage series --since/--until and rig bundle history --since say when a bound isn't a date or time.
Requests the daemon now answers cleanly: a route given a JSON body of null answers 400 invalid_json instead of 500; a limit of 2^63 or more on queue overdue and usage series answers 400; chat search behind rig ask matches _ and % literally; rig restore's not-found hint gives the working rig snapshot list <rig>; rig watchdog register warns when a cadence such as 5m will be read as 5 seconds; and rig whoami --host <id> --full forwards --full, which a remote older than v0.4.0 refuses.
Give a team environment variables in its spec: an env: block on the rig and on any member in rig.yaml reaches every seat, including Claude Code's cleared environment, carries into restores and handovers, and rig export writes it back. rig up <spec> --env KEY=VALUE adds a value for one launch. Names match ^[A-Z_][A-Z0-9_]*$, and OPENRIG_* is reserved. The values aren't secret: they're stored in OpenRig's database and snapshots and copied by export and bundles.
Members added later with rig grow can't carry a member-level env:. rig up --env is not forwarded for flat-node specs or v1 bundles, although the launch may succeed. Do not rely on the flag for those formats in this release.
Off by default: bundle.services_enabled (OPENRIG_BUNDLE_SERVICES_ENABLED). A bundle carries its declared Compose file, rig bundle inspect shows each service before anything runs, and rig up and rig bootstrap install it into a lasting target folder.
With bundle.services_enabled on, rig up <file>.rigbundle installs into the current folder when you don't give --target.
rig workflow instantiate and rig workflow run take --project, --mission, --slice and --tags, stamped on every packet so the work shows on the slice's Queue tab; instantiate-lifecycle takes them from the lifecycle. An unknown project creates the instance unscoped with a warning.
A workflow's preferred_targets entry can be written member@{rig} to address whichever rig the instance is bound to. An edited workflow spec restored with an older modification time is picked up.
rig bundle create refuses harness login files (.claude/.credentials.json, .codex/auth.json, .pi/agent/auth.json) anywhere in an agent folder or carried project. An existing archive holding one without a record now fails verify, unpack and install; install answers 400 even with --force. rig bundle create --include-login-files includes those files only when recipients should have them and the login's owner agrees, and records their paths as included_login_files.
rig scope mission create writes status: planned, which keeps a new mission out of the project's proof rollup until you set it active. With --workspace <project> it creates the mission inside that project, under its declared missions.root, which now wins over a missions/ folder for every command that reads missions.
The execution view binds a queue row to a mission by its exact mission: tag first, and a slice: tag can name a slice by its SPEC id: as well as its folder.
The execution view shows a slice parked on an external: gate as a waiting lane, and shows a merged:<pr> report on the merged rung, counting it as merged only with a candidate:<sha> that matches the slice's built commit. A slice dependency can name a sibling by its SPEC id:.
rig bootstrap and rig requirements resolve a relative spec path from your own folder when the target is this install's local daemon.
Files saved with Windows (CRLF) line endings read the same as Unix ones in SPEC.md intents, checkbox blocks, PROGRESS.md, proof and review headers and artifact-pool watchdog reads.
The Scopes view reads a proof drop's folded evidence references and mini-requirements numbered 1). The daemon's scope audit answers "mission not found" for a mission name that leads outside the slices root.
A member's own mechanic survives bundling and other spec rewrites.
A blank list item in a rig spec, agent spec or startup files: list gets a validation error at its path instead of a crash, and an agent spec profile written as a bare key reads as an empty profile.
Context packs and agent images accept file names with two dots inside a name, such as notes..md; a real .. segment is still refused.
An empty Docker Compose ps result reads as no services instead of unparseable output. On macOS, rig file copy reports how many bytes it copied.
For a Codex user, the rigs skill makes one optional offer of a removable rules file so six rig commands run without approval prompts.
The operator reads the keeping-a-watch page when a person asks it to watch something or remind them, and the page teaches a "warn me when my allowance runs low" watch.
The rigs skill has a person's own Claude Code or Codex offer to join the team before its first message to the operator, so the operator sees a named teammate; inside tmux it says the effects first and the person chooses.
Setup re-checks a failed network check with network access before asking anyone to sign in again, installs through rig setup, and the operator offers the Workshop configuration that fits the person's providers.
delegating-work says to brief a peer according to the model it runs, which rig ps --nodes --json shows as model. The openrig-user skill adds an opt-in loop that wakes an attached app's agent when work arrives. rig ask --wake help and two skills explain the target forms.
User-visible text says "harness" for the agent software around a model, and "community harness" for one an outside contributor maintains; docs/reference/topology-naming.md defines the terms.
The README points contributors' agents to the developing-openrig skill, the roadmap links the public board, and the contributor guides describe the one-home skill workflow.
Daemon environment, off by default: OPENRIG_TMUX_CONTROL=1 sends the daemon's routine tmux commands through one long-lived control client.
OpenRig 0.6.12 adds rig community-harness list for installed community-harness presets, showing their maintainer and declared limits. These harnesses are not tested by OpenRig. Support is launch-only: no resume or fork, no resource projection or startup-file delivery, and identity remains unconfirmed. A quiet screen is not proof that the seat is idle.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. Twelve database migrations run on the first start, whatever you use: 100 (the typing guard's mode columns), 104 (nodes.effort_choice), 105 (two usage_samples columns for Claude's cost estimate), 107 and 108 (Pi usage entries and charges), 109 (the mode store's unique index per category, keeping every row), 110 (workflow_instances.scope_json), 111 (rigs.env_json and nodes.env_json), 112 (two indexes for recent queue activity, about 10 to 12 MB on a long-running install), 113 (applied_launch_observations.daemon_opt_out), 114 (sessions.agent_app) and 115 (the Slack thread map rebuilt by channel and thread timestamp, keeping existing rows). There are no migrations 102 or 106, and no earlier migration was changed.
A runtime's --version check gives up after 30 seconds. At daemon start a hung probe no longer keeps the daemon from listening, and rig bootstrap now blocks on a runtime that is installed but fails or times out its check, which it reports apart from one that's missing. rig up's preflight fails a Claude, Codex or OMP check that doesn't answer within 30 seconds (a warning for Pi), and stopping the daemon ends the probes it started.
On Linux and macOS, the daemon runs tmux directly instead of through /bin/sh. Set OPENRIG_TMUX_SHELL=1 when starting the daemon to keep the old path.
The daemon does much less work on a busy machine: it runs tmux without a shell, captures up to 24 transcripts in one tmux call, skips reading terminals that haven't changed, and reads queue activity and obligations with far fewer queries. The timings are the authors' own.
The daemon now declares string-width (^8.3.0), which was already in the installed tree, so the lockfile adds no package. The daemon package exports two more entry points, ./composer-prompts and ./workflow-scope. The Node requirement is unchanged (node >=22). Install scripts are unchanged: the package's only lifecycle script is still postinstall: node scripts/check-abi.mjs.
A daemon with its own OPENRIG_HOME but the real HOME or CODEX_HOME writes one advisory line to daemon.log and starts anyway.
Getting started pins the rigs skill to your installed release, with separate Claude Code and Codex commands and a Linux note for Codex's sandbox.
The README says only the older web UI is in maintenance mode.
The Linux checks run on pushes to main; shipped skills regenerate from one home in the repository; a daemon-restart scenario joins the installed-package checks; and the snapshot tick looks up its restore source only when a seat changed.
A custom permission policy saved with Windows (CRLF) line endings now takes effect. It used to fail to parse, so the seat fell back to the advisory floor; now it resolves to its declared posture at the next launch or restore.
The temporary launch script removes itself with /bin/rm or, if that's missing, the shell's own rm, so the agent still launches under set -e.
OPENRIG_HOME counts as unset. A community contribution.With HOME unset, rig auth and the daemon look for Codex sign-in files in the account's real home. An exported but empty OPENRIG_HOME is treated as unset by the TUI's socket, the terminal token lookup and rig crash-cart.
Tests use stub runtimes, clean up their temporary folders, keep compaction packets out of the shared folder and pass on macOS's temporary folder.
Timing-sensitive tests wait on conditions; the native transcript suite runs concurrently again; cross-host usage gains tests; and two repository scripts were fixed.
rig commands can reach the first user's OpenRig. In the first-user check, the second person's ordinary commands reached the first person's daemon on its default port (7433), read one of its queue rows and sent a request to the first person's operator, which ran its checks and refused to work on the other user's private project. Running each user's daemon on its own port let both people work, but it doesn't keep them apart. The code involved matches 0.6.11's; 0.6.11 wasn't run to compare. It's being diagnosed for 0.6.13.partially_restored: the three seats meant to be running came back, and a deliberately stopped seat was left awaiting a decision.submitted, unconfirmed as reached, and confirm by the restore request or the drop in context (#1483).rig policy current and plain rig ps don't show a policy stored with rig policy apply --seat; rig seat status and pod-aware export do (#1285).env: and --env values aren't secret, and members added later with rig grow can't carry a member-level env: (#1294).rig up --env is not forwarded for flat-node specs or v1 bundles, although the launch may succeed. Do not rely on the flag for those formats in this release (#1294).complete before the new Claude is ready, because the readiness check can match the previous occupant's text still in the pane, so the handover packet may be typed before the successor starts. 0.6.11 has the same behaviour. Until it's corrected, check that the successor read its packet (#1544).rig slack verify on the hub doesn't check relayed channels, and an answer from a seat other than the one the message reached still posts at the top level (#1436).unknown with usage windows on, until a seat's reading can be linked to its login (#1504).openrig-user skill's subscription section mentions apps that don't ship in this release; don't look for them (#1450).bundle.services_enabled on, rig up <file>.rigbundle installs into the current folder when you don't give --target (#1474).rig seat restart --keep-conversation supports Linux only, and its native result hasn't been re-run on 0.6.12.blocked because of one seat that isn't running: it folds a restore forecast into current status (#1141).rig scope slice create says it couldn't verify membership, the slice was kept. Check mission.yaml before creating it again (#1007).rig doctor checks both harnesses; rig tui --shared attaches on the default tmux server; a team's first steps can still ask for approval; a seat waiting at a permission prompt shows no product notice of its own; a launched team writes into its working folder; a resumed Claude seat can't be confirmed after /clear, an in-process /resume or a hand-started process; Codex team seats don't ask before lifecycle commands; the send gate trusts the latest recorded hook; and after a restart rig status shows Kernel: skipped.claude alias defined only in a nushell rc doesn't resolve; after a reboot, start the daemon with rig daemon start and bring your rig back with rig up <name> --existing; the web UI's file preview runs scripts, so preview only files you trust; browser access isn't complete browser isolation; and Codex network inside the sandbox hasn't been checked with real logins, managed policy bundles or on Linux.A parked seat whose occupant is known isn't woken again for the same rows after a wake was delivered, until it updates one of them or one changes state. A failed, refused or unrecorded wake still repeats at the next park, and a new occupant gets its first wake. "Delivered" means the wake was typed into the seat's terminal, not that the agent read it.
rig send --wait-for-idle N waits before it takes the seat's delivery lease, so other sends and queue wakes to that seat aren't held behind a long wait. N is now a deadline from the start of the wait, including any time spent waiting for the lease. A timeout names the wait you asked for.
Restore no longer starts a seat fresh when an earlier occupant left a resume token. If a seat wasn't running when the snapshot was taken, but an earlier session recorded a resume token, rig up --existing and rig launch stop that seat at awaiting-decision and start nothing, because the token doesn't prove that conversation can still resume. rig launch of such a seat now exits 1 where it used to start it fresh. The message names the command that starts it fresh on purpose. For a seat stopped when the snapshot was taken, restore does not resume an earlier recorded conversation; the offered --fresh command starts a new conversation, not that recorded one.
The status API for a rig, GET /api/rigs/:id/status, can call a running rig blocked because of one such seat: it folds the restore forecast into the rig's status, though no seat is halted. rig status and the TUI's rig list don't show it for a running rig, and the web UI's "Resolve & restore" still works.
A Codex seat at its usage limit stays unknown with reason usage_limit while its screen redraws, instead of reading running, and rig ps --nodes shows usage_limit under REASON. It clears when a readable screen shows no banner, or when Codex shows it's working again.
A Codex seat can keep reading as at its usage limit when an old banner is followed by a submitted prompt and the agent's reply, with no Codex working line and no input line below. The case was constructed from the source and hasn't been seen in the real sessions reviewed.
Restart a Codex seat in place, on Linux. rig seat restart <seat> --keep-conversation --reason "<why>" restarts one Codex seat in its own pane on the same conversation, so it starts again on whatever account its Codex home now holds. It doesn't switch accounts itself. Before it stops anything, it checks that the seat is a Codex seat OpenRig launched, in a single pane, with a conversation it can prove, and that relaunching wouldn't change the running settings; any doubt refuses with nothing stopped. --plan runs every check and stops nothing. --verify --expect-account <account> reports moved only when a tool turn has completed in the same conversation since the restart and Codex's status names that account.
Linux only: on macOS, a request that gets as far as reading the running process refuses before stopping the seat. It has restarted one real Codex seat (Codex 0.153.4), doesn't verify which account the seat resumes on, and wasn't tried with Codex builds that share one app server.
rig up warns on stderr when it chose / as the launch folder itself.
rig seat launch --fresh --stop checks before it stops anything. A seat with a still-live older session is refused before the current agent is stopped; 0.6.10 stopped it first. --plan previews a fresh launch without changing anything and exits 0 even when it would be blocked. --reason is required only to execute; leaving it out reports missing_reason.
A seat you stop on purpose no longer makes its rig read degraded. rig seat stop records the stop, rig ps and the TUI leave that seat out of the rig's state, and running counts still exclude it. rig ps --nodes --json carries intentionallyStopped. A rig whose seats were all stopped this way reads stopped.
In a restore of a rig whose seats are in pods, a seat is rolled back as unresumable only when its runtime reports the conversation missing. When Claude, Codex or Pi reports that, the restore rolls the seat back to awaiting-decision and starts nothing, instead of reporting a failed launch and leaving the new pane behind. A rig where every seat ends this way reads partially_restored instead of failed; the exit code is 1 either way. A Claude that exits to its shell for another reason keeps 0.6.10's result: the restore reports the seat failed and keeps its pane, so you can read what happened. Restores of rigs without pods keep their earlier handling.
Seats no longer inherit the launching Claude Code or Herdr session's identity variables, such as CLAUDECODE. If removing them fails, the launch goes ahead with a warning.
A late readiness result no longer overwrites a newer launch attempt.
receipt: line instead of the whole task.New switch, off by default: output.compact (or OPENRIG_OUTPUT_COMPACT). The seven queue write commands (rig queue create, claim, unclaim, update, block, handoff and handoff-and-complete) print a short receipt instead of the whole row, and --full prints the whole row. While it's off, their output is unchanged; --full is accepted, and rig config lists the key.
rig queue create --verify to a registered person who is also a seat with a pane reports the Slack receipt beside the terminal wake, in a gateway object (posted, transport-failed, never-posted or still-pending). Posted isn't read. GET /api/queue/:qitemId returns ownerNotificationLevel.
The wait is bounded by the existing 30-second default, and it can use all of it even when Slack isn't set up or the row won't be posted; the Slack side then reads still-pending. The 30-second wait wasn't tested on the exact package.
rig queue whoami --work-candidates now returns up to 50 held and 50 next rows, with heldTruncated and nextTruncated; heldAndNextTruncated now means either list was cut. With OPENRIG_REFOCUS_WORK_PACKET on (off by default, as in 0.6.10), the refocus hook's work packet lists held and next work separately, up to 50 each.
rig queue create --verify reports outcome: "claimed" for a claimed row whose wake wasn't confirmed, instead of indeterminate. This was a known issue in 0.6.10.
rig send --verify confirms enveloped multi-line messages that the agent's screen indents, and queue wakes use the same check.
A handed-off closure with no visible successor gets the same custody advisory and stuck-sweep finding as a done closure.
The chat send and topic routes return 400 before storing anything for payloads that used to fail with 500.
/api/wake-resolve answers a fractional or non-numeric generation with its usual "does not exist" reply.
Watchdog registration refuses a scan_interval_seconds that can't be turned into a number, such as non-numeric text, a boolean or an empty array.
A forked Claude seat (rig fork, or session_source: fork) launches with its own session ID, so strict restore and clear-attention can verify it as its own conversation.
Forking a Claude seat needs Claude Code 2.0.73 or later, which OpenRig doesn't check. Forks launched before this keep reading mismatch until they're relaunched.
New switch, off by default: provider.usage_windows (or OPENRIG_PROVIDER_USAGE_WINDOWS in the daemon's environment). rig provider status --refresh reads a Codex login's usage windows once. While it's off, --refresh says how to turn it on and exits 1.
rig whoami now shows a Codex agent's conversation ID once it's confirmed resumable. A community contribution, toward a community member's request.rig whoami --full --json shows a Codex seat's runtimeContext.threadId when its stored conversation ID was verified resumable.
A failed Pi or OMP turn shows the provider's own reason when it gives one.
A Slack reply sent with "Also send to channel" in the default channel is recovered after a connection gap.
The TUI says when a queue sample hit its cap or failed to load: counts read N+, a seat missing from the sample shows ?, and Pulse no longer lists a NEEDS YOU item a second time.
The caps count the whole host, so a quiet rig can read ? while another rig is busy.
The TUI Topology table's Q, WORK and NOW columns fill again on drilled views and on a seat's page. They had been empty there since 0.5.14.
The what-changed note, capability map and changelog were written for 0.6.11.
rig ps takes a rig's running count and status from the same set of sessions as rig ps --nodes.
Discovery of unmanaged panes matches agent process names exactly, plus claude.exe and names cut to 15 or 16 characters.
A member's cwd in a rig spec can be ~, start with ~/, or use $VAR or ${VAR}, expanded from the daemon's environment, when the path as written doesn't exist.
rig proof show and rig proof judge take --project <id> or a <project>: scope prefix, resolved through the workspace catalog. Scope readiness for a project selected in the TUI uses that project's own root.
Parallel rig scope slice create calls get different slice numbers. A number is never reused, and each create leaves a hidden slices/.nn-NN claim file.
After rig scope slice create says it couldn't verify membership, the slice was kept. Check mission.yaml before creating it again: creating it again makes a second slice with the same name.
A reference page on keeping a watch for a person with today's queue commands.
The page says a timer fires once, but a held-back wake uses up that fire: when the typing guard holds a one-shot --wake-after wake back, the timer isn't retired and can wake the owner again at each interval. Read the task's transitions after each wake.
The rigs and openrig-skills skill descriptions split by who is asking: a person in their own session, or an agent inside a rig.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. One database migration runs on the first start: 103 adds one column to sessions, intentionally_stopped, which is 0 for every existing row, so OpenRig can tell a seat you stopped on purpose from one that exited. There are no migrations 100 or 102. The bundled openrig-core plugin stays at 0.1.9; two of its files changed.
No npm dependencies were added, removed or changed, and the Node requirement is unchanged (node >=22). OpenRig's own version fields advance to 0.6.11. Install scripts are unchanged: the package's only lifecycle script is still postinstall: node scripts/check-abi.mjs. The one database change is migration 103.
rig config set daemon.port refuses a port outside 0–65535, and a port like that from the environment or config.json falls back to the file or default with a warning. rig daemon start --port refuses one too, with a clear message and exit 1.
Daemon and CLI tests that failed on some Linux machines (ext4 inode reuse, dash as /bin/sh, a 002 checkout umask) pass there now.
The pane-isolation tests clean up after themselves, and the TUI default socket-path test no longer fails on macOS.
The portability report runs when it's started through a symlinked path.
rig setup --providers claude, codex or both.4 added13 changed26 fixedSetup asks only about what you use. rig setup --providers claude (or codex, both, or a comma list) installs and checks only the harnesses you chose. One you didn't choose reads skipped, never failed. Before, a person who used only Claude Code saw Codex checks fail. The first hour also stops showing tmux's "error connecting to …" when the daemon starts, and the next steps say the remaining steps are part of the install, not extras.
rig doctor still checks both harnesses, including one you don't use. rig setup --providers doesn't change it.
A resumed Claude seat whose session ID changed now proves. Claude can continue --resume under a new conversation ID. When the first SessionStart hook from the process OpenRig launched reports that resume, OpenRig records the new ID. The seat restores as resumed, and sends stop warning that its identity can't be confirmed. Before, such a seat stayed attention_required.
A resumed Claude seat still can't be confirmed after /clear, an in-process /resume, a hand-started process or a launcher shim, or with a plugin older than 0.1.9.
Codex seats behind a launcher: a seat whose codex is a launcher that starts Codex as a child now receives rig send and queue wakes. Before, every send was refused with "reports sh as the foreground command". This had been broken since 0.5.17.
Attach commands name the right tmux server. If the daemon was started inside tmux on a non-default socket, every attach command it prints now carries -L <name> or -S <path>. From inside tmux, run it as env -u TMUX <command>. This was 0.6.9's known issue about attach commands reporting "no sessions".
rig tui --shared still attaches on the default tmux server, so it can't find the dashboard when the daemon runs on a non-default tmux socket. Use rig tui there instead.
rig launch no longer calls an exited agent running. On a seat whose agent exited to a shell in its kept pane, rig launch <rigId> <seat> returns attention_required and exits 1, with the ways back. It used to report the seat already running.
rig up <rig> --existing reports a seat still missing part of its guidance. The rig then reads partially_restored instead of fully_restored, and the command exits 1.
rig up --existing puts back the OpenRig guidance that rig down removed from CLAUDE.md, CLAUDE.local.md or AGENTS.md. This happens before a resumed seat starts, and the person's own text isn't touched. Before, the seat came back without its role and culture guidance while the restore read fully restored.
A restore no longer trusts a launcher's word. A launcher that isn't Claude can name one conversation while the Claude it started runs another. That restore now reads attention_required instead of a false resume.
A seat whose tmux session is confirmed gone reads detached, and rig ps --nodes says to restore or relaunch it. Before, it read as needing attention and could still look active.
rig start says why it skipped an existing kernel and how to bring it back.
New escalation setting: queue.wake_human_rung defaults to always. Set it to explicit-only to keep agent-to-agent escalations off the person's Slack unless the row is addressed to a person.
Compact rig queue list rows carry summary, and so do overdue and undelivered. The fields they leave out are now absent instead of null or "", and fieldsElided names them.
lastNudgeResult reads claimed on a claimed row whose wake wasn't confirmed. The recorded value is kept in the new lastNudgeWireResult.
rig chatroom history returns the newest messages when you give neither --after nor --since, still listed oldest first. Before, a long room returned its oldest.
rig queue create --verify reports an agent seat's wake result and any claim as soon as one is recorded. Before, it waited 30 seconds and said still-pending.
It can report outcome: "indeterminate" for a row the seat already claimed. Read claim for pickup: neither the claim nor this outcome shows the message was displayed or read. It isn't a failed handoff, so don't re-run the create.
_ and % in a rig view show <lens> --rig <name> rig name are no longer wildcards, in all seven lenses that filter by rig.
rig ask matches chat keywords such as config.json or C++ literally.
rig heartbeat reads queue files with Windows line endings, and picks its latest proof correctly on a host not set to UTC.
The queue's recovery lookups no longer scan every row.
Claude's faint suggested prompt in an empty input box no longer reads as a person's draft, so a send to that idle seat goes through. Real typed text is still protected.
A Codex seat keeps its last context reading across quota-only usage records.
A fresh Codex launch keeps git metadata paths that end in whitespace.
A Slack decision click that isn't recorded now gets one reply in the thread saying why. It covers clicks from someone who isn't registered, clicks on a decision sent to someone else or no longer waiting, and buttons that no longer match. A recorded click still gets its usual reply. It's best effort and needs the Slack bot token.
The TUI counts CR+LF as one Enter, keeps an emoji whole at a wrap, pages Local reading directories by 10, and skips fenced examples when opening a file at a heading.
rig add takes a rig's name as well as its ID.
rig capture --lines N shows the last N lines. When earlier lines are cut, stderr says how many. In --json, lines is then the count returned, and requestedLines and omittedLines are new. For the old output, N history lines plus the whole pane, add --history-plus-pane, which also keeps the old JSON without those two fields. Without --lines, nothing changed.
When a caller of the daemon's terminal API names no provider, the daemon uses cmux if Herdr isn't available, and says so. rig terminal open without --provider, and the TUI's "Open terminals", open Herdr, or plain tmux when Herdr is missing, never cmux. The TUI's preview now asks for the same provider as the window it opens, so it isn't refused when cmux is running.
The what-changed note, capability map and changelog were written for 0.6.10.
A tmux started inside a Herdr pane is recognised by Herdr's own markers.
A ?limit= that isn't a number returns 400, not 500.
rig usage top|series --json prints the daemon's error as JSON, and rig usage top reads every sample in its window.
rig workflow instantiate and project print the instance and packet they made.
rig transcript --tail keeps a line ending in %.
--json written after -- no longer turns an error into JSON. A community contribution.--json after -- no longer turns an error into JSON.
rig env prints the daemon's error instead of a JavaScript one.
A file read over 1 MiB ends at a whole UTF-8 character.
Scope and progress: progress rows go only under a matching ## heading, frontmatter updates stay inside the exact --- delimiters, a capability delta expires only with a readable successor file, and fenced examples no longer count as slice sections or progress tasks.
rig package install looks for a guidance block's closing marker only after its opening one.
Teams are guided to ask less and say more. Starter and factory guidance now directs teams toward first steps and checks that need fewer approvals. The operator is instructed to explain what a launch changes in your project folder and to offer a separate copy before launch. Operators and leads are instructed to surface a waiting seat's prompt and attach command.
The first-user check still needed four team approvals and didn't surface the reviewer's prompt until the person asked.
Tests and repository checks: a scenario for the rig up round trip; the docs guard keeps unusual file names exact; the portability report handles header-shaped added text; and GET /api/bootstrap lists same-second runs newest first.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. Two database migrations run on the first start, and neither changes existing data: 099 adds one partial index to the queue table, so the queue's recovery and stuck-sweep lookups stop scanning every row, and 101 adds four empty columns to sessions, so OpenRig can remember a resumed Claude conversation's new ID. There is no migration 100. The bundled openrig-core plugin goes from 0.1.8 to 0.1.9.
No npm dependencies were added, removed or changed, and the Node requirement is unchanged (node >=22). OpenRig's own version fields advance to 0.6.10. Install scripts are unchanged: the package's only lifecycle script is still postinstall: node scripts/check-abi.mjs.
On WSL2, a first rig daemon start after a timed-out probe goes ahead when the port is free. It's refused only when the port is really held, and the refusal names --port.
A typed Slack reply to a decision with structured questions is now kept on the row, in humanAnswers. Before, the decision closed with no recorded answer. The reply is placed under the first unanswered question and marked as typed, so text that matches an option's ID is never read as clicking it. The decision still closes, and closing doesn't mean approval.
A cross-host rig send --host to an http-registered host has a 30-second request budget, plus any --wait-for-idle time, instead of 5 seconds. If the answer doesn't come back, or the connection drops after the request went out, the send reads "Delivery UNCONFIRMED" (remote-outcome-unknown, exit 1) and tells you to check the target before any resend. Only a host that was never reached reads as unreachable.
OpenRig now notices a Slack Socket Mode connection that stays open while events stop, from Slack's server pings and from the echo of OpenRig's own posts, and replaces it: the new connection opens first, at most once every 5 minutes. rig slack status adds a Delivery: line beside "connected", and the connections view says whether delivery is confirmed.
The refocus work packet is in this release, off by default: nothing changes unless it's enabled.
A kernel whose first start failed and whose seats then recovered, for example after rig seat continue, now reads ready in rig status, with a separate line naming the earlier failure. Before, it kept reporting bootstrap_failed, and its agent list could show a seat twice.
A long rig chatroom wait keeps waiting through a slow or restarting daemon instead of ending at the first failed poll: poll timeouts, connection refusals, and reset or closed connections are retried until --timeout. The first failure prints one line on stderr; other errors, such as a removed rig, still end the wait.
Claude Code seats installed from nixpkgs are recognised as Claude and read running. Before, every one read attention_required. A seat that already carries an attention marker from an earlier failed launch keeps it until rig seat clear-attention.
rig slack status shows how many retained records for inbound messages, reactions and click answers are waiting to be retried; it counts records, not distinct messages. The backlog is kept across restarts, and if it can't be read, status says the count is unknown and why.
A Slack alert whose first post failed is kept and retried when the gateway next starts; if its row has closed by then, the alert is no longer posted, and the row says why. Alerts for rows still open, decision-resolved notices and digests still post, and rig slack enable says how many earlier posts wait to be retried.
The what-changed note, capability map and changelog were written for 0.6.9, and they now say rig launch takes a rig ID, not a rig name.
rig scope <tier> approve --workspace <path> now stamps the work tree you named. Before, it reported success while the stamp landed in the daemon's own work tree. If the named workspace has no missions/ folder, approve now refuses instead of stamping the configured one.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. There is one database migration: 098 adds an index on (qitem_id, transition_id) to the queue's transition history, and a 0.6.8 daemon still opens the upgraded database. The bundled openrig-core plugin stays at 0.1.8.
No npm dependencies were added, removed or changed, and the Node requirement is unchanged (node >=22). OpenRig's own version fields in the package files and the lockfile advance to 0.6.9. Install scripts are unchanged: the package's only lifecycle script is still postinstall: node scripts/check-abi.mjs.
rig launch <rig> <seat> --plan now previews a single agent's launch, as --seats already could: the snapshot it would use, the target, and what happens to the other agents, without launching anything. It works over --host too.
One skill being edited in the managed skill catalog no longer blocks every agent's skills. Before, any uncommitted or untracked file in the catalog made it unavailable, so rig context work-install, rig up, launch and restore preflight failed for every agent. Now that skill is skipped and named in a warning, the remaining clean skills stay available, and an agent's launch only warns. An agent that already has a copy of the skipped skill keeps it as it was, and the next applied projection can refresh it once the change is committed or restored. No agent receives uncommitted files.
An uncommitted change to catalog.yaml itself still makes the whole catalog unavailable, because it changes what every selection means.
When a person replies inside a Slack thread OpenRig opened, an agent's answer with rig queue create --human-intent update --reply-to <that reply's task> now posts in the same thread instead of as a new message. The messaging-the-human skill says how.
Not tested with live Slack.
rig launch --plan reads the daemon's version first and refuses to send to a daemon older than 0.5.9, which would ignore the plan and launch.
If the answer to rig launch --plan isn't a plan but shows the daemon acted, the command exits non-zero and tells you to check rig ps --nodes -A before retrying.
After a managed /compact, a Claude agent that takes no turn is still restored: the turn boundary, the restore request and the read-depth audit now reach it. Before, they waited for the agent's next turn, so an agent with nothing to do could sit unrestored until something woke it. Each step still waits until the agent's screen shows it idle, and only a fresh usage reading can start a compaction.
Not tested with a real compaction.
rig down still saves its recovery snapshot when finding one agent's resume details fails, instead of skipping the snapshot for every agent.
When an agent is woken for a task handed to it, the reminder now shows the task's summary on one line, up to 120 characters, so it sees what the task is for.
A Claude agent launched with --remote-control confirms its identity instead of staying "identity not confirmed".
When Slack rate-limits a post and asks for a pause of 10 seconds or less, OpenRig waits that long and retries the post once. A longer pause is still left to the usual retry of retained messages.
Not tested with live Slack.
The check that keeps the Slack bot token on Slack's own hosts now sits inside the private-file download itself, not only at its one caller.
On an empty TUI command line, j and k move the selection down and up.
Not tested in a live TUI session.
TUI shortcuts that could swallow a typed command moved to capitals: the footer toggle from f to F, and in a selected Scopes view the mini-requirements and narrative keys from m and n to M and N. Agents drive the TUI by typing, so commands such as find, feed, mission, narrative and needs now reach the command line whole.
Not tested in a live TUI session.
A topology naming reference explains how to name a team's pods and agents, from purpose to domain to role, and the rig spec, bundle and authoring guides point to it.
rig workflow validate now reports step_cannot_finish, and refuses to instantiate the workflow, when no allowed exit can finish a step or route onward. An allowed mapped exit is valid, and in a dependency graph handoff can finish a last step without a next one. Before, such a workflow passed validation and then stalled on that step.
A spec that validated on 0.6.7 can be refused now. The message names the step and the fix.
rig scope audit accepts slice folders numbered 100 and above, which rig scope slice create makes after 99.
rig proof add --file keeps a leading UTF-8 byte-order mark in the artifact instead of removing it.
rig context work-install --runtime <runtime> and rig skill loadout --runtime <runtime> exit 1 when a selected skill is skipped for uncommitted catalog content. They only inspect by default; applying with --apply-skills or --apply still reconciles the remaining clean skills.
Exit 1 alone doesn't tell you whether files were changed.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. The running daemon keeps its old build until it restarts. This release has no database migrations, and the bundled openrig-core plugin stays at version 0.1.8.
Building from source: the CLI and TUI build scripts now finish their output with one Node script, scripts/prepare-workspace-build.mjs, instead of chmod, mkdir and cp, so they no longer need POSIX shell tools. No npm dependencies or install scripts changed.
This hasn't been run on Windows yet. WSL2 is still the way to run OpenRig there.
The architecture and CLI reference docs were brought up to date for 0.6.7.
The README, reference guides and architecture docs were brought up to date for 0.6.8, and the what-changed note, capability map and changelog written for it.
The rigs skill carries the whole route from install to the operator: when the install is finished, how to hand the operator your goal, folder and branch, how to relay the operator's questions to you, and what to give you when no window can open.
After your agent installs or starts OpenRig, it now opens the OpenRig TUI and the kernel's operator for you as the next step, with rig terminal open saved:kernel --window, instead of waiting for you to ask. The same command brings your agents back later.
Checked on a test Mac, where the view opened without being asked within 15 seconds of the handoff after install. The full first install from nothing, Linux desktop windows, opening the view from Claude Desktop and the plain-tmux layout without herdr weren't tested.
A message or watchdog reminder no longer answers a question for you. When a Claude agent is waiting at a question menu or an approval that OpenRig can't read on screen, a rig send message or a watchdog's reminder is refused if the agent's latest hook said it was waiting for you, and it goes up the usual escalation path instead. Before, the message's Enter could pick the first option.
The check trusts the latest recorded hook. If an agent's hooks are recorded out of order, a message can be refused for a question that's already answered, or let through to one that's waiting.
An agent on an upgraded install can read what behaves differently, and what to do, with rig context get reference/whats-new.md (or $OPENRIG_HOME/reference/whats-new.md with no daemon running). The openrig-upgrade skill ends by reading it and telling you what affects you.
rig doctor now tells you whether Claude Code and Codex are installed and signed in, the same way setup does. A pass means a credential is on your machine, not that the provider accepts it, and it fails for a harness you don't use.rig doctor checks the Claude Code and Codex installs, and a login or configured provider credential for each, the way rig setup does. It exits 1 when either is missing or has neither, including a harness you don't use, and its output and --json gain claude_install, claude_auth, codex_install and codex_auth rows.
A pass means a login or configured provider credential is available locally, not that a provider accepts the credential or that an agent can work. It fails for a harness you don't use; only the harnesses your teams use need a login.
agents.advisor_session now defaults to the kernel's advisor agent, which every install has, and rig restore-packet write without flags targets the calling agent's own rig.
The daemon itself moves a finished restore map into place before compacting, so the agent no longer needs a shell rename that could stop at a permission prompt.
rig restore-check tells you what to do. A restore can still report a problem with an agent that works.A running agent whose current pane fails its identity check reads attention_required. The stored startup result moves to storedStartupStatus, and rig restore-check names the verdict and what to do.
A restore can still report a problem with an agent that works (#273), as in 0.6.6.
After a daemon restart, an agent waiting at an approval keeps showing as needing attention.
rig seat handover <seat> works on the same agent again and again, and running it again recovers an agent left half handed over.
An agent taking over from a Codex predecessor now sees its messages in the recap.
After a managed Claude compaction, the agent's refocus context arrives with the restore request instead of being lost to the acknowledgement prompt, also when the request arrives wrapped as pasted text. If refocus fails three times in a row for an agent, one issue appears on the stream (rig stream list --tag refocus). A per-agent refocus health file is kept under $OPENRIG_HOME/refocus/.
Checked on a fresh test install; it hasn't been proven on a long-running install.
rig up, instead of being refused for lack of a snapshot. A community contribution.rig up restores an agent that was launched fresh and then stopped with a clean rig down, instead of refusing with no_snapshot.
A very long snapshot history no longer breaks retention.
The built-in starter and factory leads publish their team's roster at first start, so rig roster find <topic> knows those teams.
A periodic reminder first fires one interval after you register it, not immediately.
Claiming a blocked task clears its blocker, and rig queue update <id> --state blocked without --blocked-on blocks it again on the previous one.
When a workflow step that was sent back completes again, the steps that depend on it run again, including ones that had passed.
rig workflow validate, instantiate and run use the file in your current folder.
rig chatroom wait --timeout exits 1 at its deadline instead of printing a late message.
rig gateway human now say that OpenRig supports one configured human.rig gateway human messages say OpenRig supports one configured human.
A waiting task's timer shows when it will really fire, never as a 1970 date.
An invalid --wake-after, such as 7d, prints an error instead of exiting silently; use hours, such as 168h.
rig workflow resume --occurrence with an ID that matches no unresolved failure is refused instead of rerunning a different one.
rig workflow run and watch exit 3 for an aborted run instead of waiting forever.
rig chatroom history --since with a value it can't read fails instead of printing "No messages.", and the error shows control characters as text.
rig chatroom watch keeps messages in order while switching from history to live.
Watchdog history keeps the right entries at tied timestamps.
At a Claude agent's next launch, its activity hook points at $OPENRIG_HOME/state/claude-activity-hooks/ instead of a script inside the project.
In Claude Code team agents, help on lifecycle commands, such as rig down --help and rig bundle install -h, runs without a prompt, and common spellings of allowed commands match too: quoted or absolute paths, env, command and exec, and a project's vitest and jest. Real lifecycle actions still ask. It works through a command hook in the agents' launch settings, and no settings file is written.
Checked in a live Claude session for only one selected lifecycle-help command; source and parser coverage is broader. Codex team agents don't yet ask before lifecycle commands such as rig up and rig down; Claude Code team agents do.
At a Claude agent's next launch, OpenRig writes its status-line command in .claude/settings.local.json with quoted paths.
OpenRig reads a Claude agent's session from its process correctly when permission settings are passed inline, so a running agent isn't misread as needing attention. This fixes 0.6.6's known issue #921.
The resume token is quoted when OpenRig types a resume command.
The operator explains a Codex agent that needs a newer Codex and offers to update it.
The starter's Codex reviewer still needs a Codex newer than 0.145, as in 0.6.6.
Managed Codex launches pass -c check_for_update_on_startup=false, so a Codex agent doesn't stop at Codex's update offer.
Updating Codex is left to you or the operator.
A sign-in or incompatible-client notice that appears at the last moment of a Codex restore is reported as needing attention with the screen's last lines, and a late "no saved session" retries fresh.
OpenRig finds a Codex agent's conversation when its home path contains spaces.
Launch preflight reports the Pi version against the tested baseline, and a Pi agent says at start whether it has a stored sign-in or provider key, and how to provide one.
Pi agents weren't tested beyond automated tests.
Pi agents on minimax/<id> models can receive MINIMAX_API_KEY when it's in recovery.provider_auth_env_allowlist.
Pi agents weren't tested beyond automated tests.
rig slack verify checks the app is in each one. Replies are matched by timestamp alone, private channels aren't supported, and live Slack wasn't tested.rig slack channel-map set <rig-or-seat> <channel>, list and remove send a rig's or an agent's posts to its own Slack channel. An agent's entry wins over its rig's, and anything unmapped uses the default channel. rig slack verify checks the app is in every mapped channel, and the TUI shows the map.
Replies are matched to a thread by its timestamp alone, so two mapped threads with the same timestamp in different channels could route a reply to the wrong agent. Missed-message recovery covers the default channel only, and private channels and direct messages aren't supported targets. Live Slack wasn't tested.
An emoji reaction on any message OpenRig posted for an agent's question in Slack now reaches the agent that asked, as a task saying who reacted and with what.
It doesn't answer or close the question. An existing Slack app needs the reactions:read scope and the reaction_added event added, then a reinstall; rig slack verify warns while the scope is missing but doesn't check the event. A reaction on a reply in the question's thread, rather than on OpenRig's own messages, is ignored. Live Slack wasn't tested.
A question too long for one Slack message is posted as numbered replies in its thread. If it still can't be posted, the agent gets a task saying so, instead of the question silently not arriving.
Live Slack wasn't tested.
A Slack reply sent with "Also send to #channel" ticked now reaches the agent that asked.
If Slack's connection stalls while reconnecting, the daemon retries instead of going quiet.
The view opens where your agent runs. Inside herdr it opens as a focused openrig kernel space, in Terminal as a new window that copies the size of your front window when it can, and in Ghostty as a new tab. From Claude Desktop, iTerm or VS Code on a Mac it opens a new Ghostty (1.3 or newer) or Terminal window at that app's default size, and macOS may ask you to allow that.
Opening the view from Claude Desktop hasn't been proven on a real machine.
On a Linux desktop, the view opens in GNOME Terminal, Konsole or xterm and asks for a readable size.
Linux desktop windows weren't tested.
While presenting the first-team choices, the operator shows each team's graph in the shared TUI.
rig terminal open now opens a new desktop window by default. If you'd rather add your agents as tiles in a herdr or cmux workspace you already have open, name it with --provider.rig terminal open <view> opens a desktop window by default. To add tiles to a herdr or cmux workspace that's already open, pass --provider herdr or --provider cmux.
If you have your own herdr config, OpenRig uses it unchanged. Only when you have none does it write a private .openrig-herdr-<hash>.toml in herdr's config folder.
The default saved:kernel view's layout follows your terminal's width. Under 120 columns the operator fills the first tab; from 120 columns the dashboard and the operator share it; the advisor always has its own tab. A kernel view you saved yourself still wins.
When no window can open, over SSH, in CI or with no display, rig terminal open says why and prints the exact command to run instead.
In the TUI, the Open terminals action uses the same launcher, works without herdr, and shows the attach command for each agent when no window can open.
Opening the same herdr view again reuses the open workspace, and your existing windows are never resized.
rig terminal open waits up to 45 seconds for a larger view to finish opening instead of reporting a timeout at 5.
In the TUI, a rig, host, pod or agent address works from any page.
A mission can declare arrangement.source.integration_ref in mission.yaml, so rig view execution checks its work against that ref instead of main. A ref that's invalid or can't be resolved reads INDETERMINATE.
rig down --host exits 2 when the remote teardown reports errors. A community contribution.rig down --host exits 2 when the remote teardown reports errors.
rig down shows one line per agent with what it was last doing, so you can see what you're interrupting.Human-mode rig down prints one line with each agent's last known activity before stopping.
Help examples and reference guides no longer use retired team names.
Agent, node, pod and agent-image names containing % are looked up correctly.
Discovery and capture keep tmux session names containing | or edge spaces, and OpenRig reads an agent's launch settings from exactly its own tmux session.
rig usage top keeps agents that reused a name apart. A community contribution.rig usage top no longer mixes up agents that reused a name.
rig usage series --since/--until compare times as moments, not text.
rig transcript --tail and --grep read a consistent copy while the file is being replaced.
rig scope copes with unusual folder paths, and a failed attempt leaves the mission unchanged. A community contribution.rig scope slice move, ship and close work when the workspace path starts with -, don't change a mission's membership when they fail, and check the folder they work on by its literal path.
rig scope mission create and slice create copy your text exactly, and frontmatter updates keep keys, comments and quoting as written.
A bare name with --mission resolves to the right piece of work past an unrelated folder of the same name.
--name still overrides it.rig bundle create, and rig up <GitHub link>, name a bundle after the name in its rig.yaml instead of my-bundle or github-bundle. --name still wins.
rig package validate, plan and install resolve relative paths, and the default --target ., from the folder you run them in, not the daemon's.
rig bundle check finds a problem, it tells you which file it's in.rig bundle check names the file in each finding.
openrig-registry repository.The docs name openrig-registry as the place to submit a team.
Reinstalling over a stopped team backs up and replaces conflicting files only in the folder the team was installed into. Installing into another folder gets first-install behaviour, refusing conflicting files before anything is written.
Teams installed before 0.6.7 have no recorded folder until their next bundle install.
A bundle install refused for a conflicting file returns HTTP 400 with a code, and other known refusals carry theirs with 409.
rig spec validate reports malformed pods, members and edges as ordinary errors.
rig bundle install --skip-version-check --force still runs the bundle's safety check.
The daemon now copies two skills for your own Claude Code and Codex sessions into ~/.claude/skills and ~/.agents/skills: rigs, which takes an agent from installing OpenRig to a team doing the work, and refocusing. They're refreshed on every upgrade.
A rigs copy you added earlier with npx skills add has no OpenRig marker, so it's left alone and not refreshed; remove it if you want OpenRig to keep it current.
The capability map agents read to learn what OpenRig can do is refreshed for 0.6.6 and 0.6.7.
The handover and compaction skills teach agents to write their recap with rig context recap-write before a handover or compaction.
rig context add --git retrieves only the current commit by default, falling back to a full clone with a warning when the server refuses.
rig skill audit passes on a clean install, and its JSON output gains new fields.rig skill audit passes on a clean install, with new --json fields.
In rig context get <pack>/<file>#section, an inline triple-backtick span no longer hides later headings, and an empty heading ends the section above it.
.git/info/exclude, so Git ignores them.Files OpenRig generates under .codex/plugins/shared:openrig-core/ are added to its marked block in the repository's .git/info/exclude.
Guidance text containing $$ or similar stays as written when OpenRig rewrites its managed blocks.
A plugin skill copy whose remembered source can no longer be projected is kept, with a plugin_skill_kept warning, instead of failing the launch; an agent selecting that plugin refreshes it at its next launch.
A project declared at / no longer ties with a deeper project when OpenRig picks one from your working folder.
The README, the getting-started guide and the help guide show a one-command install with a stable link, and a --dry-run preview that prints its four steps and changes nothing. If only provider sign-ins are left, the install itself finished. The script is the repository's scripts/install.sh, which isn't in the npm package; its closing and plan text changed too.
The full first install from nothing wasn't tested.
If permission prompts are slowing you down, the operator and the guides point to the workshop team and say plainly what access it takes.
The getting-started guide covers kernel agents that fail at first start on a tmux build that crashes, with a check and a recovery.
WSL2 is the way to run OpenRig on Windows: the getting-started guide records one user's working setup and both ways a Pi agent can get credentials.
Windows and WSL2 weren't tested in automated tests.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. The running daemon keeps its old build until it restarts. On its first start it runs two database migrations, one recording each team's install folder and one mapping replies in a Slack thread back to the question an agent asked there, and installs the updated bundled openrig-core plugin, which running agents pick up when they next launch.
rig setup now installs herdr for the view of your agents, by running herdr's own installer. Pass --no-herdr if you don't want a third-party installer run; if it fails, setup warns and you get plain tmux.rig setup installs herdr by default on macOS and Linux, for the view of your agents, by running herdr's own installer (curl -fsSL https://herdr.dev/install.sh | sh) into $HERDR_INSTALL_DIR or ~/.local/bin. Pass --no-herdr to decline it. If the install fails, setup warns and the view uses plain tmux.
This runs a third-party installer by default. The plain-tmux layout without herdr hasn't been proven on a real machine.
--ghostty. Setup no longer installs cmux, and a cmux you already have keeps working. Scripts reading setup's JSON should look for the new herdr and Ghostty steps.On a Mac, rig setup offers Ghostty once and installs it with Homebrew (brew install --cask ghostty) only when you pass --ghostty; --no-ghostty declines. Setup no longer installs or configures cmux, and an existing cmux keeps working. In rig setup --json, the cmux_install step is gone and herdr_install and ghostty_install appear.
CLI help, setup's next steps and onboarding point "show me my agents" and "the welcome screen" to rig terminal open saved:kernel --window.
If the agent installing OpenRig has a tool call refused by its own permission rules, the guides tell it to show you the refusal rather than report a failure.
The kernel's operator runs routine inspection (cd, pwd, echo, wc, sort, python, command -v and WebFetch) without asking, reads /rigs pages with WebFetch, and runs its first-boot checks as separate commands. The kernel's Claude agents get the launch-settings command hook too, with a wider allowance, and no settings file is written.
In a live Claude session the operator still hit one prompt in nine calls.
Setup's permission question now matches the team default: it recommends keeping it and offers extra remembered allowances only if you want them.
For "stop everything", the operator and the getting-started guide give the order: each rig with rig down, kernel last, then you run rig daemon stop.
The reference guides were brought to the 0.6.6 release, the architecture docs restamped to it and CHANGELOG.md gained its 0.6.6 section; the README and reference guides were then brought to 0.6.7.
With the web UI on, a terminal view that fails to attach closes that viewer instead of stopping the daemon.
After you install OpenRig, the kernel's operator agent greets you first and asks, once, what you want to build or change. It offers three teams, starter, workshop and factory, with one recommendation and a drawing of each.
This is guidance the agents follow, not a scripted wizard; how closely an agent follows it is what OpenRig's first-user runs check. In the first-user test on a Linux VPS the person reached the operator and the goal reached the team's lead, but they were guided to attach a terminal five times, which counts as help. The final package wasn't put through another journey.
See a team before you install it. rig bundle inspect, and the view printed before rig bundle install and rig up <bundle>, show the team and models, its permission settings including whether prompts are off for every agent, files the agents receive, startup actions, writes, outside domains, the author's setup preconditions and what stays unknown until launch. Facts about one agent are labelled with that agent.
The author's setup preconditions are shown, never run, and some facts stay unknown until launch.
rig up, and rig bundle create, inspect and install, accept a public GitHub folder link (https://github.com/<owner>/<repo>/tree/<ref>/<folder>). The ref resolves to one commit, and the result reports the source, configuration ID and package digest.
Link import needs git on your PATH and a running local daemon. It runs your git with credential helpers, hooks and global config disabled.
The operator fits the team you pick to the providers you have by writing an adapted copy, shows you the plan and launches only on your yes. It then shows you the team and hands your goal to the team's lead as a task, so the lead starts from that goal instead of asking again.
Guidance for the agents, checked by first-user runs. In the first-user test the person was guided to attach a terminal five times, which counts as help.
Claude Code agents in your teams need fewer permission prompts. OpenRig passes an allow list and an ask list at launch, so rig commands, reads inside the project and common test commands such as npm test, pytest, go test and cargo test run without a prompt, while lifecycle commands such as rig up, rig down, rig bundle install, rig seat stop and rig daemon stop still ask.
Applies outside the kernel at an agent's next launch, resume, fork or handover, only when no member or rig permission_policy and no saved permission choice is set; permission_policy: none keeps the old behaviour. These are command allowances, not containment: a project's test command runs project code. Your own ask and deny rules still apply, and no settings file is written. Asking before lifecycle commands works in Claude Code team agents only; Codex team agents don't ask yet. In a Claude Code team agent, rig down --help also asks; rig help down prints the same help without a prompt. rig setup's permission question still describes the behaviour before these defaults; its text is corrected in the next release.
--non-interruptive on rig up and rig bundle install lets agents whose policy is full bypass skip Claude Code's and Codex's first-launch warnings through launch flags. The choice is saved on the rig and kept for restores, forks and handovers, and rig config set launch.non_interruptive true makes it the default for new rigs.
Only agents whose policy is full bypass skip the warnings. Existing rigs start with it off. --no-non-interruptive clears it, with the rig down first. No harness settings file is written.
runtime.readiness_timeout_seconds (1 to 600, default 30) gives slow agents more time to become ready at launch, restore and handover.
Managed Claude compaction now writes its restore map inside the agent's own launch folder, at .openrig/compaction/preparation/<session>/<attempt>/RESTORE-MAP.md, with a .gitignore that keeps it out of Git.
Tools that don't read .gitignore, such as rsync, zip deploys and Docker build contexts without .dockerignore, still see the folder. The OpenRig home is used when the workspace is unknown or unwritable.
rig seat set-permissions takes --operator <address>, so it works from a plain shell, and reports a missing identity, mode or reason as three separate errors.
Startup files delivered later get the same submit check, and staged or unverified startup text shows as a warning in rig bootstrap, rig import, rig seat launch, restore records and the TUI.
An agent stopped at a provider's own prompt, such as Claude's trust dialog, is reported as needing attention instead of ready.
rig down can't stop an agent, that agent keeps its CLAUDE.md or AGENTS.md, so it isn't left running without its instructions.When rig down can't stop an agent, it leaves that agent's CLAUDE.md, CLAUDE.local.md or AGENTS.md in place instead of stripping or deleting it.
rig roster list, show and find read JSON rosters from <workspace.root>/rosters/ and show who to involve for a purpose and why, across rigs and hosts. They never assign work.
rig queue create, handoff and handoff-and-complete return once the change is saved, and the agent is notified after that. The result no longer reports whether that notification was delivered; use --verify to wait for delivery.
rig send --dangerously-interact now types your answer as keystrokes instead of a paste, so a numbered menu gets the digit you sent. This addresses a known issue, #519.
It wasn't tried against a live Claude Code or Codex menu. Upgrade the CLI and the daemon together for this.
rig queue block help says --on is always required, and blocking on a task this daemon doesn't have suggests an external gate instead.
rig up prints for you, delivers them into the same conversation without a relaunch.If a fresh Claude Code agent stops at Claude's bypass-permissions warning, it now keeps its startup context and reports attention as bypass_consent_gate. Accept the warning in that agent's terminal, then run rig seat continue <seat> to deliver the context into the same conversation without relaunching; rig up and rig bundle install print the command.
A new built-in policy, builtin:auto, launches Claude Code agents with --permission-mode auto, while Codex and Pi agents launch as they normally would. rig seat status prints the effective permission mode and where it came from, and the built-in policies are described in the RigSpec reference.
After delivering a Claude Code agent's startup text, OpenRig sends one short line asking it to run its startup check, so Claude agents finish getting set up on their own.
A Claude Code pane is read as idle when update, focus-events or weekly-limit warnings sit below an empty input, in every permission mode, and a pane showing live work alongside the mode bar is read as working.
A tall Claude Code question picker is recognised as a waiting question.
OpenRig can identify a Claude process running under a shell wrapper with a version number as its name, by its executable path, so restore attention on such agents can be cleared.
On macOS the daemon may run the built-in /usr/bin/osascript for this; whether macOS shows a permission prompt for it on a Mac that has never approved one hasn't been observed. A restore can still report a problem with an agent that works (#273); several fixes target it and it isn't fully resolved.
rig up --existing keep reporting their activity to OpenRig, and a restore that can't set that up again says so.An agent brought back with rig up --existing keeps its Claude activity hooks, and a restore that can't reapply them now says so.
Forking a Claude agent quotes the parent session ID in the launch command.
Codex agents in your teams need fewer permission prompts too. They keep the workspace-write sandbox and can also write OpenRig's workspace folder and the state folder of their pod (their group of agents), and ordinary rig, read and test commands ran without prompts in testing.
Codex team agents don't ask before lifecycle commands such as rig up and rig down yet; Claude Code team agents do. Applies only when no member or rig permission_policy, saved permission choice or named Codex profile is set; permission_policy: none keeps the old behaviour. rig setup's permission question still describes the behaviour before these defaults; its text is corrected in the next release.
Codex agents no longer stall at launch, resume or fork when Codex 0.160 shows its update menu with the new header.
An Oh My Pi agent whose model is written anthropic/…, openai/… or litellm/… now also receives that provider's base-URL variable when you allowlist it in recovery.provider_auth_env_allowlist.
If you already allowlist ANTHROPIC_BASE_URL or OPENAI_BASE_URL, those Oh My Pi agents now send requests there. Oh My Pi agents weren't tested beyond automated tests.
A rig with a Pi agent no longer fails preflight when pi isn't on the daemon's PATH; it warns, and the pane's shell finds Pi at launch. Pi flag errors name the Pi version.
Pi agents weren't tested beyond automated tests.
Pi agents keep their daemon routing and identity when shell startup files set conflicting values.
Pi agents weren't tested beyond automated tests.
A Pi agent failing with "No API key found" now says what to set.
Pi agents weren't tested beyond automated tests.
When Oh My Pi exits during startup, the error says how it exited and shows its last stderr lines.
Oh My Pi agents weren't tested beyond automated tests.
rig slack status shows the connection and whether it's catching up.Top-level Slack messages posted while the daemon was disconnected now arrive late, as tasks marked "Recovered after a gap". rig slack status shows the live connection and recovery state.
Needs Slack inbound configured. Each channel gets a recovery checkpoint on the first daemon start, and history before that checkpoint stays unknown. Live Slack recovery wasn't tested.
Slack gateway state survives a crash in the middle of a write.
rig terminal open saved:kernel works without any saved-view file: it opens the TUI, the advisor and the operator side by side, for Claude-only, Codex-only and mixed kernels. A view you saved yourself under the name kernel still wins.
Without herdr or cmux, setup, the TUI preview and a failed rig terminal open give an exact tmux or ssh -t attach command for each agent, and rig tui --shared is named as the dashboard, not the conversation.
rig up.A rig spec opened in the TUI starts on a graph of its topology, with Launch: it asks for a working folder and hands the terminal to the real rig up.
Team views say the dashboard is the overview and the lead's pane is where you talk about the work.
Shaped by an earlier first-user run on macOS, which wasn't rerun on the final package.
The TUI goes straight to the work views once it has read the daemon, even when no rig is running. The Start page is still there on S.
The query behind the TUI's recent queue activity does less sorting on large histories.
rig terminal open now uses the daemon's own tmux, and with herdr it reports a pane that closes at once as degraded instead of treating it as working.rig terminal open starts tiles with the daemon's own tmux, and with herdr reports a pane that exits at once as degraded.
rig telemetry events, transitions and tenures read one bounded page of retained history at a time, with a cursor and explicit gaps, so collectors no longer read OpenRig's database.
Requests that carry an X-OpenRig-Diagnostic-Attempt header are traced in the daemon's slow-operation log, to help diagnose inventory timeouts.
rig project shadow-drain and shadow-stop take --actor, so they run from a plain shell.
Lifecycle help says what stopping or restarting a live team does to its work.
A text change, packaged without a new journey run.
Tearing down services through the daemon's API deletes volumes only when the request says volumes: true as a JSON boolean. rig env down --volumes is unchanged.
rig host pair --timeout caps an infinite or overflowing value at about 25 days instead of waiting forever.
rig whoami can't find the agent, the error names the daemon it reached, so you can tell if it's talking to the wrong install.When rig whoami or rig health --self can't find the agent, the error names the daemon it reached.
rig project experimental refuses a FIFO instead of hanging.
rig capture now shows an empty agent screen as empty.rig capture shows an empty pane as empty.
rig mcp serve shuts down when its input ends, and keeps the daemon address it found, including IPv6.
A mission title with : , # or quotes writes valid frontmatter.
rig env logs --tail 0 now prints no earlier lines, as you asked.rig env logs --tail 0 prints no past lines.
rig env status keeps working when a cached service record is damaged.rig env status survives a corrupt cached service record.
Transcript cleanup applies runs of backspaces correctly.
A bundle can declare presets in configurations.yaml. rig bundle configurations <spec> lists them, and --preset <name> or --seat pod.member=runtime picks one.
rig bundle install --cwd <dir> sets every agent's working folder, and --target sets the install folder.
rig bundle check <folder> checks a bundle against the authoring rules, locally.
It's advisory.
rig bundle create --context-pack <dir> carries a context pack from outside the rig folder, and --project-dir <dir> carries the project the rig works in.
Installing a team that already exists shows what's installed and offers to use it, stop and replace it, or cancel. A running team is refused before anything is written; a stopped one is replaced in the folder holding its installed manifest, with conflicting files backed up first.
Reinstalling into another folder that holds a copy of the team's installed manifest treats that folder as the team's install folder, so its conflicting files are replaced. They're backed up first and the backup is named.
The v1 bundle formats are documented in docs/reference/bundle-formats.md, with JSON Schemas.
The workshop, OpenRig's team for building software in any repository, OpenRig's own included, installs from openrig.dev/rigs, which shows its install command. It has a lead, a builder, a code reviewer and a QA agent; the lead starts from your goal, plans the work and routes it to the team, and nothing is published until you say so.
It ships separately, in the openrig-world repository, needs OpenRig 0.6.6 and isn't built in, so rig up workshop doesn't resolve. Its agents run with permission prompts off and install non-interruptively by the team's own declaration; the view before install says so. The current workshop was checked separately with a startup check, not a full journey, and not in other configurations or with Pi. The first-user journey used an earlier workshop revision.
A rig spec can declare non_interruptive: true itself. An explicit flag still wins, and the declaration wins over the machine default.
Three specialist teams were renamed: adversarial-review is now code-review, research-team is now research and pm-team is now pm.
There's no alias for the old names.
rig up first-project keep working and now start the starter team, which pairs a Claude Code builder with a Codex reviewer. On a Codex-only machine, ask the kernel's operator to adapt it, and use a Codex newer than 0.145.rig up first-project still works and starts starter, with a note that the team's providers changed. first-project used two Codex agents; starter has a Claude Code builder and a Codex reviewer.
On a Codex-only machine, ask the kernel's operator to adapt it. The starter's Codex reviewer needs a Codex newer than 0.145, which refuses its model with "requires a newer version of Codex"; Codex 0.160.1 works.
conveyor, demo, implementation-pair, product-team, first-project-claude, first-project-mixed and factory-rsi are no longer built in. Rigs you already created from them keep restoring.
A new rig with services gets a Docker Compose project named from its rig ID instead of its name, so two rigs whose names sanitize to the same string no longer share containers. Existing rigs keep their project.
A rig created fresh, not as a same-name replacement, won't attach to containers or volumes under the old name-based project unless you set services.project_name.
rig up refuses before starting anything.If service boot or every agent launch fails, a rig's new Compose project is brought down, never with its volumes. When the earlier rigs being replaced use different projects and no project_name is set, rig up refuses with compose_project_conflict and starts no services, and rig down exits 2 when it keeps a project that a live rig of the same name still uses.
rig specs preview starter and rig specs preview factory end with the three-team choice.
Skills, plugins, workflow specs, context packs and agent images in a bundle are put in place before any agent launches, and routing failures are printed instead of dropped.
Installing a bundle never overwrites or merges into an installed context pack with the same name; it reports already_installed or kept_existing.
rig bundle install, so both ways of installing a team behave alike.rig bootstrap <file>.rigbundle now goes through the normal bundle install.
The built-in teams have new names. The shelf is now starter (a Claude Code builder and a Codex reviewer), factory (a lead, an advisor, a builder, QA, design and two reviewers), the specialist teams code-review, research and pm, plus secrets-manager and the kernel. Shipped help, skills and onboarding use the new names.
Every rig bundle create records the configuration ID and the assembling OpenRig version in bundle.yaml, and bundles built by an installed OpenRig record its real version.
rig bundle inspect refuses the same unsafe archive entries that install refuses.
A legacy bundle whose project block escapes its root is refused.
Installing a package whose manifest gives a skill or agent a name that would write outside the install target is refused before anything is written.
A project can list world packs under install.worlds.
The rigs skill gives an ordinary Claude Code or Codex session a route into OpenRig. It lives in the GitHub repository rather than the package; add it with npx skills add mvschwarz/openrig --skill rigs.
The implementer and QA agents no longer load the test-driven-development skill by default. The skill is still available.
CLAUDE.md or AGENTS.md shows in git status, and OpenRig prints the exact ignore line if you want it hidden. Only its Codex plugin files are ignored automatically, in a marked block you can edit.Files a launch creates now show in git status. When a launch creates a new guidance file such as AGENTS.md or CLAUDE.md in a Git repository, OpenRig leaves it visible to Git and prints the exact line you can add to .git/info/exclude. New files under .codex/plugins/openrig-core/ are excluded automatically, in a marked OpenRig block in info/exclude where deleting a line makes that path visible again.
At launch, OpenRig runs your git to check generated files and may append its marked block to info/exclude.
With several projects declared and no --project, rig context work-install picks the project whose rigs: list names the agent's rig, then the deepest project root containing the working folder, then the only unclaimed project, and reports how it chose. It stops with project_required only when more than one still matches, printing a ready-to-run command for each candidate.
The getting-started guide and the agent help guide explain rig seat continue, startup attention, readiness timeouts, Claude's bypass warning and builtin:auto.
rig context add starts a stopped local daemon.
The bundled skills index sends agents that are changing OpenRig itself to the developing-openrig skill.
The developer documentation was refreshed, with each claim checked against the code at the time: the README, CONTRIBUTING, SUPPORT and SECURITY; the reference pages for specs, projects, bundles, operations and process; the architecture pages and CLI reference; and ARCHITECTURE, DESIGN and the TUI README.
A final pass against the release commit is a separate change.
The contributor skill asks whether an agent with instructions could do a behaviour before adding code for it, and points to the capability map first.
When an agent's profile selects a plugin, such as openrig-core, its skills are now copied into the agent's .claude/skills/ or .agents/skills/ folder, where Claude Code and Codex actually load them. Before, neither runtime saw them.
Checked with stand-in harnesses; whether a real Claude Code or Codex agent lists and loads each skill is checked separately. Existing agents get the skills when they're next instantiated, not on restore. An existing folder with the same name that OpenRig doesn't own is kept, with a plugin_skill_kept warning.
rig mode effective now use the projects you configured and pick a task's project the same way project context does, so they agree on which project a task belongs to.rig mode effective, the attention view and the health readout now read the project catalog you configured, and when a task names only its mission they pick its project the same way rig context work-install does.
When a profile selects a skill from its own agent folder that differs from the managed catalog, the launch uses the selected skill with a skill_bundle_precedence warning instead of failing.
rig context sync finds a workspace context-pack folder created after the daemon started.
Context packs with .mjs or .py helper files install from the CLI.
rig compact-plan names the compaction restore skill that does.rig compact-plan points to the shipped claude-compaction-restore skill, and shipped skills no longer point to skills that don't ship.
For continuing work, the operator records one mission, and one piece of work under it, in OpenRig's workspace rather than in your repository. A question just gets an answer.
Guidance for the agents, checked by first-user runs.
rig setup prints how to reach the operator after ready, incomplete and dry-run results.
After installing, the agent opens a new terminal space with the kernel's TUI, advisor and operator, using herdr, then cmux, then a plain terminal window, without taking over its own terminal.
This is guidance for the installing agent. In the first-user test the person was still guided to attach a terminal five times, which counts as help.
The repository has an optional scripts/install.sh that checks for Node and npm, runs npm install -g @openrig/cli, runs the installed Node.js and SQLite check, then rig setup --dry-run and rig setup, and stops at the first failed step.
It isn't part of the npm package and isn't documented yet.
Restart the daemon after upgrading the CLI: run rig daemon stop, then rig daemon start. The running daemon keeps its old build until it restarts, and on its first start it installs the updated bundled openrig-core plugin, which running agents pick up when they next launch.
Agents in the built-in kernel rig launch with wider permissions by default, so its operator can install and run teams for you. Claude Code runs in acceptEdits with an allow list for file tools, reads under your home folder and operational commands such as rig, tmux, npm, git, ssh and curl; Codex runs unsandboxed with approvals off. rig seat status reports the source as the kernel operational default.
Applies only when no explicit permission choice, member or rig policy, or (for Codex) named profile is set; set one to keep the previous behaviour. Your ask and deny rules still apply to Claude Code.
An agent that installs OpenRig for you now keeps the kernel running and passes your goal and project folder to the operator, instead of building the project itself.
Part of the guidance for an agent finishing an install is on openrig.dev's install page, not in the package.
When the operator offers the workshop, it reads the workshop's current pinned version from the registry's raw file, because a cached copy can be older.
A text change, packaged without a new journey run. That a fresh first-user run then finds the current workshop hasn't been reproduced.
rig doctor no longer gives a blanket all-clear, so a clean run isn't mistaken for a working setup: it points you to its warnings and says it doesn't check provider logins or agent readiness.rig doctor no longer ends with "System checks look good." It says there were no failures, to review WARN and SKIP results, and that provider logins and agent readiness aren't checked.
In rig setup, the guided next steps now come before the permission-policy menu, and --json output gains nextSteps.
When rig setup finishes with problems, it lists each failed step with its fix.
rig daemon start --no-kernel and rig start say the operator is skipped too, and how to start it later.
Setup's next steps offer the kernel view before you pick a team, and treat declining, SSH and headless use as valid.
The README and getting-started say a Linux distribution's own Node.js can be too old and how to install 22 or 24, that npm 11 may skip the postinstall check, and that cmux is optional and macOS-only.
Contributor setup uses npm ci.
The kernel's advisor agent no longer assumes you're on macOS.
On Linux, rig setup --dry-run reports the Homebrew and cmux steps as skipped.
A rig spec can declare runtime: omp to run Oh My Pi seats, each with its own home, config, skills and sessions.
No live authenticated session against a real provider was run. rig seat clear-attention can't yet recover an OMP seat after a full restore; use rig up --existing <rig> --fresh <seat> instead.
A decision sent to a person in Slack can carry one to four questions, each shown as a row of buttons. A typed reply in the thread still works.
A Slack app created from an older manifest must turn on Interactivity by hand, or the buttons do nothing. Tested against a simulated Slack, not a live workspace.
Managed Claude compaction now asks the seat to write a restore map first, and waits for it before sending /compact, up to 25 minutes for automatic compaction. If no map arrives, that attempt stops instead of compacting an unprepared seat.
Not tested against a real Claude compaction. When no map arrives, automatic compaction stays off for that seat until you run rig compact again.
Each seat can have its own reasoning effort. Put effort on a pod member, in a profile's preferences or in an agent's defaults, and OpenRig passes it to Claude as --effort and to Codex as model_reasoning_effort on fresh launch, resume and fork.
OpenRig doesn't check the value, and Pi seats ignore it.
/clear. Shut the team down at night, bring it back in the morning, and each agent returns to the conversation it was really in.After /clear, rig down and rig up now bring a Claude seat back to its current conversation, not the one from before the clear. At shutdown OpenRig reads the live Claude process's session and saves its current ID.
Shutdown can take slightly longer. A conversation change in the moment between that read and shutdown hasn't been verified.
Activity, identity and structural sweeps read tmux in a few batched calls instead of one or more per seat, so large fleets stay responsive.
rig restore-check tells you whether a rig would really come back, because it runs the same checks a restore runs, against the snapshot a restore would use.rig restore-check now checks the real queue store and each seat's selected Claude hooks, and runs the same pre-checks a restore would run against the snapshot it would use. A rig spec it can't locate reads as yellow, not checked, instead of red.
rig down no longer forgets a rig whose agents might still be running. It says what it couldn't confirm and keeps the record.rig down keeps a rig it can't confirm is gone. If tmux can't be reached, it says so, keeps the rig record and blocks --delete.
Ending a seat's session detaches any terminal viewing it, instead of switching that terminal to another session.
Restore skips damaged snapshots, and picks the newest of snapshots or checkpoints saved in the same second.
dev and dev-2, can't be mixed up when one of them stops, because OpenRig addresses tmux sessions by exact name.OpenRig addresses tmux sessions by exact name, so a seat whose session ended can't be confused with another whose name starts the same way.
A seat that becomes ready after about 20 seconds now starts within the 30-second readiness allowance.
rig queue create picks the item's ID before sending and prints it, so after a timeout you can check that exact item and retry with the same ID instead of creating a duplicate. A same-ID retry with a different body warns that the new body wasn't saved.
A queue write to a seat that doesn't exist in a known local rig still goes through, with a warning that names the likely typo.
rig queue create --source works from a plain shell outside a seat, recorded as a self-declared source.
extended-keys is on.Typed and pasted text is submitted with tmux's named Enter key, so it no longer sits unsubmitted when extended-keys is on.
Work blocked behind a prompt is escalated as one alert to the orchestrator or operator, without typing into the blocked seat, and its wake retries resume once the prompt clears.
A watchdog attached to a parked queue row survives the other row's claim, completion or re-park.
OpenRig confirms a managed Claude seat behind a shell wrapper or a native install from the native Claude process and its exact saved conversation ID, in restore, readiness and identity checks and in rig seat clear-attention. Acknowledging a seat's attention no longer reports its conversation as resumed without strict proof.
OpenRig keeps your own Claude status line instead of replacing a custom command in .claude/settings.local.json, and no longer wipes that file when it doesn't parse.
A seat that keeps its own status line doesn't run OpenRig's collector, so its context usage reads as unknown and resume-token capture is skipped.
rig send and queue wakes reach a Claude seat whose launcher shim starts the real Claude binary.
OpenRig presses Enter once more when a Claude startup prompt is still visibly staged, and starts the seat with a warning instead of failing the launch.
When both Claude usage limits are used up, provider usage reports the later reset time.
Each OpenRig install can have its own Codex home. Set an absolute CODEX_HOME when you start its daemon, and managed resume uses that home; with it unset, installs still share ~/.codex.
Before upgrading, align a CODEX_HOME set in your shell rc with the daemon's, or a Codex seat may not find an earlier conversation.
Codex seats can now reach the local daemon under OpenRig's default workspace-write sandbox, so they can do their queue work. OpenRig turns on network inside the sandbox only when you haven't made your own choice and no managed requirement restricts it.
Not checked with real logins, managed policy bundles or on Linux. To keep it off, set network_access = false under [sandbox_workspace_write] in your Codex config.
Codex 0.157 and 0.158 seats showing the empty-composer placeholder are recognized as idle, and a busy seat whose status row sits far above the composer isn't misread as idle.
Long Pi launch, fork and resume commands run from a temporary script instead of being cut off by the terminal's line limit.
A follow-up can go into an earlier Slack thread, and video and PDF evidence files up to 50 MiB are attached.
Tested against a simulated Slack, not a live workspace.
Slack sweeps skip building expensive queue views for items with nothing to post, and each seat's latest usage sample is found through an index instead of by sorting its history. Both are based on korallis and Lee's work in korallis/agent-stack.
On a synthetic 200-row queue the Slack sweep was much faster when few items needed a post, and about 11% slower when none had been posted.
Slack messages in different channels with the same timestamp are both delivered, and retries search up to 10 pages of history for an earlier post with an unknown outcome.
A second TUI on the same OpenRig home binds its own control socket and shows its path, instead of taking over the first one's.
The TUI's live activity updates back off when the stream keeps dropping and retry when it's slow to open, and arrow keys work in application-cursor mode.
rig ps --resources now shows what that watching costs your machine, and idle agents are watched less often.rig ps --resources shows load per CPU, running seats and what transcript capture costs your host. Idle panes are captured less often, and changes to transcripts.lines and transcripts.poll_interval_seconds reach running seats without a daemon restart.
An idle seat's transcript can trail its pane by about 6 seconds. Load average isn't CPU use.
Commands that used to print an empty or successful result when the daemon refused the request now exit 1 or 2, including rig bootstrap, rig env status, rig plugin list, rig context list, rig specs ls and rig start --all.
rig health gives a partial answer for the busiest families instead of failing, and names any source it couldn't read.rig health above 200 families evaluates the 200 busiest and marks the result PARTIAL instead of failing, and a failing health source is named as unavailable while the others still report.
Execution views no longer freeze the daemon while their Git checks run, and a repeated session-boundary line no longer grows a transcript on every capture.
CLI commands, the daemon's own endpoints and rig host list work with an IPv6 daemon host.
rig status and similar commands say it didn't respond, instead of telling you it's stopped.rig status, rig start and four other commands say when the daemon didn't respond, instead of calling it stopped.
rig ps counts agents stuck at a permission prompt in each rig's attention total, so you can see who's waiting for someone.rig ps counts a seat stuck at a pane-only permission prompt in the rig's attention total.
rig usage top shows missing usage data as unknown, so it isn't mistaken for an agent that used nothing.rig usage top lists seats with missing counters as unknown instead of zero.
rig scope refuses a stale mission edit instead of silently dropping another edit, and refuses a slice that depends on itself.
rig daemon stop finishes promptly even while something is still listening to its live events.rig daemon stop ends open event streams after a short grace period instead of timing out on them.
rig spec validate runs locally, with the same validators the daemon uses, so you can check a spec without a running daemon.
rig spec preflight still needs the daemon.
Agent specs can point at a plugin with openrig-home:plugins/<name>, and built-in agents load openrig-core from a custom OPENRIG_HOME.
Service readiness accepts bracketed IPv6 targets and requires every Docker Compose replica to be healthy.
Agents learn the route to your project's context. The onboarding text and bundled skills teach rig context work-install, which lists what the current project declares: its intent, context files and skills.
Contributors get architecture maps, a developing-openrig skill and a roadmap.
OPENRIG_URL or PATH somewhere else. One less way for a seat to end up on the wrong install without anyone noticing.Your shell startup files no longer change a seat's OpenRig settings. Each launch re-applies the daemon's OPENRIG_URL, OPENRIG_HOME and related settings, and puts the rig that matches the running daemon first on PATH.
For classic Claude launches, a claude alias or function defined only in a nushell rc doesn't resolve, and neither do some older fish setups.
The web UI is off by default. The CLI and the TUI are the supported ways to use OpenRig; the web UI is in maintenance mode.
To turn it back on, run rig config set ui.enabled true, then stop and start the daemon.
After a full down and up, a managed Claude seat resumed in auto mode takes ordinary messages again once OpenRig confirms the same Claude process is running in its pane. A resume it can't confirm asks for your attention instead of starting a fresh conversation.
A restore can still report a problem with a seat that works.
A blocked queue row keeps its park timer through a seat handover, and a rerouted row's periodic timer is retired.
Codex readiness works with providers other than OpenAI.
The onboarding text new seats read now covers both ways a build goes wrong, and the skills router says a project, rig or machine may add its own skills.
An agent runtime OpenRig can't verify is reported separately from a stopped agent.
A rig's only seat can be relaunched fresh with rig seat launch --fresh --stop, even after stopping it ended the tmux server.
If the resume check can't tell whether an older Claude seat's conversation came back, the restore keeps the new session and asks for attention instead of closing it.
A seat launch that times out reports an unknown outcome instead of a failure.
Claude's capability check, run before a managed launch with an explicit permission mode, gets five seconds instead of one, which helps when several seats start at once.
A timeout can still refuse the launch.
OpenRig recognizes a native Claude install behind its managed shell wrapper, including versioned install paths. When it can't observe the runtime, ordinary delivery goes ahead with a warning.
Removing Codex activity hooks leaves the rest of your Codex config untouched.
Slack keeps a long message from a registered person whole, where anything past 1,800 characters used to be lost, and a storage error during retries is logged instead of escaping.
Checked with synthetic connections on Linux, not a live Slack test.
A daemon started with rig daemon start --port is now the one your other local commands talk to.
rig proof add --replace swaps in a new file, so a symlinked or hard-linked artifact no longer overwrites the other path's contents.
CLI request deadlines cover the whole response, and a body that fails after its headers is reported as an unknown outcome.
SSH, rsync and Herdr output keeps UTF-8 intact across chunks.
Atomic file edits keep the executable bit.
rig env status tells you when a service's last result is out of date, so you don't take it as current.rig env status says when a service receipt is stale.
Importing YAML over a stopped rig with the same name archives the old one and shows its ID and the command to unarchive it.
rig bundle create keeps binary files in agent packages intact instead of re-encoding them.
The daemon checks which address and which web page a request comes from. Nothing changes for the CLI, the TUI, agents, the queue, Slack or other OpenRig hosts; a custom DNS name or proxy domain goes in OPENRIG_ALLOWED_HOSTS, and a local app on another port in OPENRIG_ALLOWED_ORIGINS.
It isn't complete browser isolation, so keep the daemon on loopback or your tailnet. Both settings need a daemon restart.
Seats launch after an upgrade removes the old version, using the running version's built-in startup files.
Native process checks no longer depend on your locale.
rig proof add --media, and existing proof isn't overwritten without --replace.rig proof add refuses binary files given as artifact text and asks for --replace before overwriting. Attach screenshots and other binary evidence with --media.
A newly created snapshot helper is cleaned up when its input target can't be established and its ownership can still be proven, so unusable helpers stop piling up.
Archived rigs no longer interfere with seat references, and removing a stale duplicate node keeps another live seat's session and queue work.
rig send reaches a running Claude Code seat launched with an explicit permission mode, even when OpenRig's managed launch leaves a shell wrapper in front of it. A bare shell alone is still refused.
The process checks used synthetic process metadata; an installed Claude send on every platform wasn't demonstrated.
Three first-project recipes give you a small owner and checker team: two Codex agents (first-project), two Claude Code agents (first-project-claude), or a Claude owner with a Codex checker (first-project-mixed). Use the accounts you already have.
Version-2 bundles copy their verified contents into --target and launch from there, so relative working directories and agent references survive after extraction. Without --target, rig up <file>.rigbundle installs into your current directory.
Message delivery won't type a message into a bare shell when an agent runtime should be running there, so it isn't run as shell commands.
ANTHROPIC_BASE_URL can be forwarded through configured provider-auth environment forwarding, so Claude seats can use an Anthropic gateway.
Known Claude and Codex runtime hints are passed to Herdr, and a missing fresh startup context or adapter is reported as needing attention.
Managed launches after a tmux or host restart keep the named launch target instead of confusing it with an old pane binding.
Actual reboot and power-loss recovery were outside this fix's verification.
Identity and transcript polling no longer start overlapping work on every tick when a sweep runs slowly.
This prevents piled-up overlap; it isn't a measured overall CPU reduction.
A sender qualified with this daemon's own host ID is recognized as a local seat, avoiding false ownership refusals.
Numeric Claude rate-limit reset timestamps are accepted.
Messages and queue nudges reach a ready Codex seat launched through a managed shell wrapper instead of being refused as a bare shell.
TERM=dumb.Pi seats keep line editing with TERM=dumb.
tmux window fields with printable separators parse correctly, and health that wasn't requested is no longer reported as a failed read.
Agent-guided setup asks once whether to allow OpenRig commands without repeated permission prompts. Yes adds native rig command rules at project scope, or user-wide if you choose, without granting general shell access.
The scope, loading and revocation of these rules weren't yet verified in this release.
rig context get help and it works through environment checks, setup problems and permissions itself.One help path: ask your agent to read rig context get help for environment checks, setup and restart symptoms, permissions, instance layout and escalation. If rig can't run, the same guide is online at openrig.dev/help/agents.
Pull requests now run eight test jobs, including an installed queue-durability scenario with a deliberately planted lost-baton failure that a checker must catch.
It doesn't certify every platform, provider or authenticated agent workflow.
On hosts not set to UTC, identity and context-usage readers compare observations with the real start of a seat's generation.
Startup and resume checks recognize Codex's » conversation prompt, including a fresh prompt after a dismissed hook-review panel.
A Codex conversation is recognized when its header has scrolled away and a custom status line puts the model name after other fields.
Recognition only; the separate stale restore warning in #116 isn't repaired here.
Codex fresh launches work from linked Git worktrees, resolving the worktree's Git directory instead of treating its .git file as a directory.
rig view show execution reads a catalogued project's execution view when you name its --project and --mission.
It doesn't infer the project from your current directory.
The bundled Vault skill has valid frontmatter, with a check for shipped skill headers.
Choose permissions per seat: rig seat set-permissions records an audited choice for that seat's future launches, floor, full_bypass or inherit. Claude modes such as auto are accepted only when the seat's Claude advertises them.
Native enforcement across fresh launch, resume and fork wasn't verified. Status output doesn't prove a mode's effective permissions; check the running session.
Open a whole rig in Herdr from the TUI. Every running seat opens in a workspace named after the rig, up to 16 seats per tab.
Check that the rig opened as expected in your Herdr version.
A typing guard for seats where you type by hand. It pauses all automatic input to that seat and keeps messages and wakes in an outbox, which you read with rig seat held-messages.
It covers OpenRig's own input paths, not other tools writing directly to the terminal.
Pi seat activity reports carry the seat's occupant generation, and reports from an earlier or foreign generation are refused instead of being attributed to the seat.
Activity reporting from an installed Pi seat wasn't verified end to end.
An optional classifier seat can have Jev, through OpenRouter, label Rig Stream observations. It's off by default and runs only when the classifier seat runs one foreground command.
Experimental. The labels are advisory, with no demonstrated accuracy, and nothing routes on them.
Codex seats launch with --no-daemon when the installed Codex supports it, so each seat's tools run in its own app-server instead of a shared one.
rig slack manifest prints the Slack app manifest for the app you create in your own workspace, offline, before any daemon or token exists. --url gives Slack's create-app link with the manifest filled in.
Experimental. The setup steps haven't been confirmed against a real app creation.
rig scope no longer trip the readiness check when they have no metadata block.Missions and slices created with rig scope no longer fail the readiness reader when their manifest has no metadata block.
OpenRig now requires Node.js 22 or 24 and uses better-sqlite3 13. Node 20 is refused with an explanation, and your data stays where it is: the daemon reopens the same database and applies pending migrations in place.
Node 26 and odd-numbered Node versions aren't supported or tested. On Apple silicon, use Node.js 22.
The kernel starter's summary says when it's showing the library preview, explains that the runtime is chosen automatically, and points to the seats actually running.