feat(bin): require local-skill declaration for ship and scout briefs - #3
Merged
Merged
Conversation
* docs: add captain-authorized inherent red-check merge exception Keep the default red-PR ban and own one always-loaded exception in the merge-authority section: captain-explicit PR or bounded batch plus exact check, only when the failure is inherent to the selected delivery path. Yolo cannot activate it; final head and the full current check suite must be verified; other substantive failures remain non-waivable. * docs: replace narrow red-check exception with captain precedence Supersede the inherent failing-check merge exception with one always-loaded Firstmate-local rule: a current explicit concrete captain instruction overrides a conflicting Firstmate-written standing rule only within exact scope, never above platform/system/developer instructions. Keep the ordinary red-PR default and yolo boundary; point section 7 at the section 1 owner.
* fix: give validation-time captain overrides a supersession sequence The Validate section let a captain instruction that completely invalidates the work being validated keep the same task and worker, but never said how: the adjacent rule flatly bans hand-editing, committing, aborting, or restarting during an active run with no carve-out, so a worker facing full invalidation had no sanctioned path forward. Add the missing sequence: cancel through no-mistakes axi's abort command, confirm the run has stopped through axi status, recover branch ownership through axi sync's guarded recovery, only then replace the obsolete work, and validate once against the final head. The existing ban on hand-editing an active run now cross-references this sequence instead of contradicting it. * no-mistakes(review): Make validation custody recovery conditional * no-mistakes(document): Clarify validation supersession abort exception * fix: keep obsolete pipeline commits out of the superseded deliverable The review-applied fix made custody recovery conditional on branch_sync.next_action.code, but left an open gap: recovering custody settles who owns the branch, not what content ships. As written, a worker could recover an obsolete run's branch and build the replacement on top of its now-irrelevant commits instead of from the correct pre-invalidation base, carrying obsolete content into the final deliverable. Make that explicit: custody recovery settles ownership, not content, so the worker replaces obsolete work from the correct base and keeps the obsolete run's commits out of what gets validated and shipped. * no-mistakes(test): Restore minimal pre-invalidation replacement instruction * fix: dedupe redundant "replace the obsolete work" restatement Line 309 already says the worker replaces the obsolete work from the correct pre-invalidation base, excluding the obsolete commits. The closing sentence restated "replace the obsolete work" again before gating the final validation run, layering the same fact twice instead of stating it once. Trim the closing sentence to just the ownership gate and the single-run-against-final-head requirement it uniquely adds.
* fix: bind explicit --backend to exact-task authority A Herdr-backed second mate carried a prior one-task --backend tmux exception forward by analogy, so its child landed in tmux and never appeared under the second mate in Herdr. Runtime detection was correct; the authority surface was not. docs/configuration.md now owns that an explicit --backend is authorized only for that exact task. AGENTS.md and fm-spawn help point there. * no-mistakes(document): Consolidate backend selection authorization documentation
…unchenguid#1229) * fix: remove projected workspaces through Herdr's focus-preserving pane-death path Herdr 0.7.5's explicit close of a workspace-emptying last pane moves the attached client's focus to a neighbor workspace, flashing the captain's whole window and routing in-flight keystrokes to the wrong pane until Firstmate's exact-tab restore masks it 56-197 ms later. Teardown and cleanup now plan a workspace-emptying close as a focus-safe removal: verify the close empties the workspace, reposition the doomed workspace behind the focused one through the verified workspace.move transport when it sits before a non-last focused workspace, prove the pane holds one lone idle shell, and end that shell so Herdr removes the emptied workspace through its focus-preserving pane-death path. Any ambiguity or failure falls back to the plain close behind the existing restore backstop, and fm_backend_herdr_kill applies the same plan for non-projected removals. Two conditions proven on real hardware are encoded in the adapter: BSD ps reports a login shell's comm as "-zsh", and an idle shell transiently hosts a prompt helper right after a workspace.move relayout, absorbed by a bounded strict-sample settle window in the idle-shell proof, now the single owner shared with session-start cleanup. An isolated-lab regression reproduces the raw steal on 0.7.5 and proves the plan removes a doomed workspace with zero wrong-focus samples and no corrective focus; unit fixtures cover the position, edge, ambiguity, move and kill failure, escalation, and transient-helper cases. Upstream fixes (kunchenguid#1877 explicit close, kunchenguid#1912 pane death) are merged but unreleased; once released the plan degrades to a harmless reorder-then-remove. * no-mistakes(review): Confirm pane death from structured not-found responses * no-mistakes(review): Serialize Herdr kills and sample focus continuously * no-mistakes(review): Synchronize Herdr focus evidence output * no-mistakes(review): Refuse unlocked Herdr pane closes * no-mistakes(document): Correct Herdr focus-safety documentation * no-mistakes: apply CI fixes * fix: never erase a Herdr task's records while its pane survives a refused close A transient presentation-lock contention could produce a completed teardown while the exact Herdr pane stayed alive as an unowned restored shell: the kill refused the unlocked close (correctly), returned success, the warning was suppressed, and cleanup erased the task's status, turn-end, and metadata records after the isolated copy had already been returned. Teardown now acquires the named-session presentation lock before anything destructive: a contended lock refuses up front while the isolated copy, the task branch, every durable record, and the endpoint are all intact for a plain rerun, and the projected and flat close paths both run under that one held lock instead of acquiring their own. Durable records are erased only once the exact pane is confirmed gone through its structured presence; a refused, skipped, or failed close retains every record with a visible, retryable error, and after a skipped close (unresolvable lock path) only a structured pane_not_found counts as gone - unknown never does. The teardown regression drives a live contending lock holder end to end: the refusal touches nothing (no worktree return, no branch drop, no close attempt), and the retry after release returns the copy, closes the pane under the lock, and removes the records. The unconfirmed projected close now refuses with records retained, and the structured-presence gate has a strict/default unit matrix. * no-mistakes(review): Require structured pane-not-found before Herdr record removal * no-mistakes(document): Correct Herdr record-retention verification date * fix: refuse ambiguity, revalidate SIGKILL ownership, and roll back failed removals Three accepted-contract corrections from the post-CI personal review of the Herdr keep-spaces focus-flash mitigation. Ambiguous endpoint identity no longer counts as a confirmed-gone pane: a missing or malformed target refuses record removal in the structured presence gate, and teardown treats missing confirmation machinery as a refusal instead of skipping the gate, so only an exact structured pane_not_found ever erases durable task records. The pane-death SIGKILL escalation re-reads the exact pane's process information and refuses to signal unless the same shell pid still passes the strict bare-idle ownership proof, so a pid that exited and was reused by an unrelated process is never signaled; the refused escalation falls back to the plain close with the unrelated process untouched. A reposition whose removal is not confirmed no longer outlives the attempt: the emptying-close plan records the verified pre-move order and original index whenever it invokes the mover, and both close owners restore the exact original workspace order through a second verified move, under the same held session lock, before reporting the close as failed. Each defect was reproduced first: the unit matrix documented malformed identity as gone, the PID-reuse regression showed SIGKILL reaching a disowned pid, and the rollback regression showed a single unrestored move. Teardown-level regressions cover unparseable presence retention alongside the strict identity matrix. * no-mistakes(review): Require confirmed Herdr removal and resolvable teardown locks * no-mistakes(review): Enforce structured Herdr closes and teardown preflight * no-mistakes(review): Preflight explicit Herdr close confirmation helper * no-mistakes(document): Document Herdr rollback failure semantics * no-mistakes(review): Captain, harden recursive Herdr teardown safety * no-mistakes(document): Document recursive Herdr teardown evidence * fix: retain nested secondmate home when a recursive child cleanup fails Captain-decided Option A correction for nm-askuser-flash-r6, found during complete-diff rereview of the merged head. cleanup_firstmate_home_children's recursive secondmate branch called itself for a nested child's home without checking the result, then unconditionally removed that home right after. remove_firstmate_home ends in an unconditional recursive delete with no check for leftover records, so a nested secondmate whose own Herdr grandchild failed its confirmed-gone check would have its entire home - retained grandchild records included - erased by the very next line. Guard the recursive call the same way every other fallible call in this function already is: || return 1, skipping remove_firstmate_home and leaving the nested home and its records for a safe rerun. Empirically, fm-teardown.sh's set -eu already halted the script on the prior unguarded call before reaching removal (verified by hand with the guard reverted, under both this session's bash and stock macOS bash 3.2) - the reachable behavior was already correct. The explicit guard is still applied exactly as decided: it matches every sibling call site in the function, and it stops the correctness of this path depending on errexit's well-known fragility under refactors (a wrapping if/&&, or a future subshell) rather than on an explicit check. Adds a teardown-level regression building on the existing direct-child Herdr fixtures: a top-level secondmate contains a nested secondmate, whose own Herdr child's close goes unconfirmed. Proves through the public fm-teardown.sh interface that the nested home, the nested secondmate's own record, and the grandchild's metadata and status all survive, and that the top-level secondmate's record survives too. * no-mistakes(document): Document nested Herdr teardown retention
…d#1431) * fix(dispatch): prioritize quota completion runway * no-mistakes(document): Document completion-aware quota runway selection
…uid#1447) * Preserve task contract in no-mistakes intent * no-mistakes(review): Preserve complete current task contract in no-mistakes intent
…nguid#1452) * fix: centralize secondmate registry parsing * no-mistakes(review): Centralize secondmate registry binding validation * no-mistakes(review): Harden registry EOF and symlink validation * no-mistakes(review): Reject unreadable registries before parsing * no-mistakes(document): Document punctuation-safe secondmate registry validation * no-mistakes: apply CI fixes
* feat(procevent): supervise long-polling sources into durable events Firstmate had no way to wait on a blocking external process without holding a conversational turn. Add a domain-neutral process-to-event runner plus a thin adapter around the currently published `lavish-axi poll` interface: canonical physical source identity, one machine-wide owner per source, direct argv execution, and durable 0600 result capture before any event referencing it is published on the existing wake queue. No second notifier, no polling control plane, and no retry machinery. A captured result with no durable handled acknowledgement stays eligible for bounded re-announcement across any number of drains and restarts. Draining a wake before acting on it and then starting a replacement session resurfaces the same exact source and sequence, and never puts result payload text in an event line. `fm-procevent.sh handled <source-id> <sequence>` is the only thing that stops re-announcement: generation-keyed, private, path-safe, durable, and atomically idempotent, so a paired external effect gated on its first-time versus repeat report is never authorized twice. An acknowledgement is refused unless matching captured result and adapter records already exist, so a premature or mistyped call cannot suppress a future result. The source side is unchanged and still lossy: the published poll clears feedback destructively before returning it, so a result lost in that window is unrecoverable. This is never at-least-once, no-loss, or lossless, and the handled acknowledgement is not a generic exactly-once effect either - a crash between an external effect and its acknowledgement can still repeat that effect on replay. Integrate registered sources with watcher supervision, the guards, and recoverable secondmate teardown across nested homes, and cover source identity, lifecycle races, supervision, restart handling, and cleanup safety with regressions. * no-mistakes(review): Prevent Lavish prompt text from spoofing missing sessions * no-mistakes(review): Serialize publication and secure handled acknowledgements * no-mistakes(document): Document hardened process-event acknowledgement guarantees * fix(procevent): never reclaim a source whose owned group still runs A runner is its own process group leader and starts the blocking source in that group, but the claim records only the leader PID and its identity. If the leader died while the source child kept running, the missing PID was classified stale: reconciliation released the claim and started a second runner while the old blocking source was still consuming the same canonical source. For the Lavish adapter that means two destructive long polls racing on one review session, so it is not harmless process litter. It also contradicted the documented promise that ownership is never released until the whole group is gone. Ownership state now distinguishes a generation that is really gone from one whose leader crashed with its group still alive. Reconcile stops that surviving group and releases its exact generation before starting any replacement, and keeps the claim for a later cycle when it cannot prove the group stopped or another home owns it. Acquisition and `start` treat the same state as held rather than reclaimable. Signalling that group is safe precisely because only an absent leader reaches this state. A reused PID leaves the leader alive, so the identity comparison still classifies it stale or uncertain and no group signal follows, which keeps the existing PID-reuse refusal intact. Add a public-interface regression for the exact crash cut - SIGKILL only the leader, prove the child group survives, reconcile, and prove the old group is gone with no second source running - plus its counterexample that a generation with no leader and no surviving group is still reclaimed. Update the runner help, operating documentation, skill, and verification record where they described reclaim in terms of the leader alone. * no-mistakes(review): Enforce runner group ownership and detect poller overlap * no-mistakes(review): Isolate runner groups from unrelated caller processes * no-mistakes(document): Document isolated process-event runner launch * no-mistakes(lint): Suppress Perl literal ShellCheck false positive
…nchenguid#1500) * fix(bin): deliver process-event results and retire ended sources Two defects reproduced during a real Lavish adapter session. One human `Send & End` produced four captured results: the real feedback, then recurring empty ended sessions. The generic runner had no way to learn a source was finished, so every reconcile restarted a poll that returned immediately. The runner now asks the source's own adapter - `fm-procevent-<adapter>.sh terminal <result-file>` - and on exit 0 alone re-proves ownership, drops the registration, and releases its own claim under one source boundary. Terminal knowledge stays adapter-owned: for Lavish that is an ended session, a missing session, and the final feedback delivery the published poll marks with `session_ended`. An adapter with no terminal command keeps its source armed exactly as before. Capture before publication, captured-result durability, queued wake durability, bounded re-announcement, handled deduplication, one-owner ownership, and explicit idempotent retirement are all unchanged. A captured result queued its `check` wake durably, but a healthy watcher with a fresh beacon never delivered it; the result surfaced only after a manual drain. Publication happens outside the watcher (in the runner) or unconditionally (in reconcile), so the watcher had no newly actionable signal to report and never reached its rewake path. It now reports a queued-but-unsurfaced process-event record through the same actionable exit every other wake uses, deduplicated by the same `.seen-*` marker discipline the signal scan uses, so the record is always durable before it is suppressed. The durable queue remains the authority and no second notifier, poller, timer, queue, or adapter-specific wake path is added. Regressions cover both, driven end to end: an armed Lavish source against a stand-in for the published poll polls once, captures once, publishes one distinct event, and retires itself; two fixture adapters prove the terminal decision follows the adapter alone; and a real capture plus a real watcher prove one proactive wake before any drain, with no duplicate wake while the record stays queued or after it is acknowledged. * no-mistakes(review): Harden process-event retirement and proactive delivery * no-mistakes(review): Route process-event delivery through shared wake owner * no-mistakes(document): Clarify process-event delivery and retirement documentation * no-mistakes(lint): Fix ShellCheck control-flow warnings * no-mistakes(lint): Fix wake output status lint warning
* perf(ci): shard the portable serial behavior lane across runners The Behavior portable serial job ran all 69 scripts of the serial remainder on one runner. The measured serial sum on run 30725985757 was 1143762 ms (19m04s) against a 20-minute timeout, so the job intermittently reached the cap and was cancelled with every step passing. Setup is only about 7s, so the cost is entirely test wall time. Split the lane into four separate-runner shards. Each shard is still strictly serial, and separate runners mean no two of these stateful scripts ever share a machine, so the split needs no concurrency isolation proof. Assignment is longest-processing-time bin packing over measured per-script duration hints, balancing every shard to 285941 ms (~4m46s) of expected work, and the timeout tightens from 20 to 15 minutes. bin/fm-test-run.sh owns the shard count and refuses a lane whose "ofN" disagrees with it, while ci.yml derives the same count from strategy.job-total rather than a literal, so changing it in either file alone fails the lane loudly instead of leaving part of the required suite unrun. --check-coverage additionally proves the shards are non-empty, disjoint, and exactly equal to the serial lane. No test is weakened, skipped, or removed. Also replace the wall-clock sleeps in the --jobs scheduler test fixture with an explicit signal handshake between the fixtures. The old 0.5s-versus-0.05s race failed on a loaded machine; the handshake passes under sustained CPU saturation. * no-mistakes(review): Correct portable serial shard balance evidence * no-mistakes(document): Document portable serial shard evidence accurately
…henguid#1545) * fix(bin): identify harness sessions by path and report delivered wakes Two supervision faults, both reported by a contributor and both open on the default branch. Fault 1: the Stop auto-arm never claims the home. fm_harness_ancestry_pid() matched only the basename of `ps -o comm=`, and Claude Code's native installer names the per-session executable by its version (.../share/claude/versions/ 2.1.220), so that basename identifies nothing. Three real failure shapes follow: a version-named session is missed entirely and the hook exits 0 with the epoch never written (unconditional on Linux, where procps reports the kernel exec name and ignores argv[0]); a claude-named daemon that directly parents sessions wins the outermost-contiguous-claude rule ahead of the session itself; and a session that is both version-named and daemon-parented has its live lock reclaimed as stale and rewritten to the shared daemon pid, corrupting the home's ownership record. Harness identity now also reads whole components of the executable path and of argv[0], which is what both platforms still carry. Matching whole components only keeps that widening safe: bin/fm-claude-stop-autoarm.sh and ~/.claude/hooks scripts have no "claude" component. Ownership is then decided against the session's whole contiguous harness ancestry rather than one chosen pid, which is the honest form of the question the library already documents ("does the current process descend from that same harness?"). That subsumes the outermost-pid rule for Claude's nested bg-spare worker chain instead of reverting it, and lets a daemon-parented session recognize its own lock. Lock acquisition still writes the outermost pid of the run, the only pid that lives as long as the session. Fault 2: an attached arm reports a delivered cycle as FAILED. The watcher prints its one reason line to its own stdout, so only the arm that forked it can read that line; an arm that attached observes nothing but a released lock and called a completely successful cycle "cycle ended without an actionable reason". No supervision event was lost - the durable queue held it - but every harness protocol reads that line as "supervision is down" and directs a manual re-arm. The arm now resolves an unobservable close against the durable wake queue, which records every wake before the watcher prints it and whose sequence counter never rewinds, not even across a drain. A cycle the queue proves delivered a wake reports that wake and exits 0; a cycle whose records a handling turn already drained reports the delivery without inventing a reason line; only a cycle that delivered nothing is still the typed nonzero failure. Fixing it in the arm covers codex, opencode, pi, grok and kimi, not just the Claude Stop path. Regressions: tests/fm-session-lock-ancestry.test.sh pins both platforms' ps semantics behind a deterministic process table and runs the real Stop auto-arm in version-named, daemon-parented, and combined real process trees, each orphaned so the walk cannot escape the fixture. tests/fm-watch-arm.test.sh drives a real watcher and a real attached arm through a real wake. Every fault case fails on the previous code. * no-mistakes(review): Bind watcher delivery records to process identity * no-mistakes(review): Return validated watcher identity atomically * no-mistakes(review): Track watcher successors by PID and identity * no-mistakes(document): Consolidate watcher arm-cycle documentation ownership
* fix(supervision): harden Claude auto-arm failure handling * no-mistakes(review): Guarantee automatic retry after Claude auto-arm failures * no-mistakes(review): Gate attended fail-open on verified supervision failure * no-mistakes(document): Document Claude auto-arm retry and guard scope * no-mistakes: apply CI fixes * fix(supervision): make Claude fail-open progression monotonic * no-mistakes(review): Preserve auto-arm failure episodes until verified watcher recovery * no-mistakes(review): Linearize auto-arm failure progression across existing locks * no-mistakes(review): Linearize positive recovery across shared failure episode lock * no-mistakes(review): Scope Claude recovery contention to Claude guard mode * no-mistakes(document): Align supervision auto-arm documentation * no-mistakes(review): Preserve actionable wakes despite healthy successors * no-mistakes(document): Refresh supervision auto-arm documentation
…d#1563) * feat(bin): require an explicit ship delivery mode in fm-brief A ship brief's definition of done was shaped by a silent per-project registry lookup, so an adjusted brief and the task's recorded delivery could disagree and no one had to decide anything per task. fm-brief now requires --mode on ship scaffolds, validates it against the closed set, refuses the conditional no-mistakes-prod-only registry policy as a task mode, and records the choice as a fixed machine-readable "Delivery contract: mode=<mode>" line that fm-spawn can check. --mode is refused on scout and secondmate scaffolds, and --yolo is refused outright because the worker never owns approval decisions. * feat(bin): require an explicit ship delivery contract at spawn and promotion fm-spawn resolved every ship and scout task's mode and yolo from the project registry, so the delivery posture was never a per-task decision and could contradict the brief the worker was about to follow. fm-spawn now requires --mode and --yolo on ship spawns, validates both against their closed sets, and reads the brief's recorded delivery contract line and refuses a mismatch before any endpoint exists; a brief scaffolded before that line existed warns once and launches on the flag. A batch carries one shared contract that each pair still checks against its own brief. Scout and secondmate spawns refuse the flags, and a scout now records no mode or yolo at all, which teardown and the snapshot already tolerate. When the explicit mode carries less rigor than the project's standing posture, a deviation notice is printed and the spawn continues, so the registry stays advisory rather than an enforced default. fm-promote requires the same two flags, because a scout carries no posture to inherit, and writes them into the task record with the kind flip. fm-project-mode keeps its one registry parser for the mechanical consumers that have no task in hand, accepts the conditional no-mistakes-prod-only annotation and maps it to its most rigorous leg for them, and grows --raw so the deviation notice can tell a conditional policy apart from a flat mode. * docs: record the explicit per-task delivery contract AGENTS.md section 7 now owns how each ship task's mode and yolo are resolved at intake, including the surface classification for a no-mistakes-prod-only project and the unregistered-project fallback, and the project-management skill defines that conditional policy as a registration-time posture with its defaults and initialization consequences. The registry blurb, script table, and architecture section follow: the registry records the captain's standing posture, and task delivery is decided per task and passed explicitly. * test: pass ship delivery flags per call site in the Herdr launcher e2e The shared spawn helper also launches a secondmate, which refuses the flags, so the contract belongs at each ship call site rather than inside the helper. * test: pass the ship delivery contract in the secondmate suites Both suites scaffold or spawn an ordinary ship task as the control case for a secondmate assertion, so each needs the explicit contract the ship path now requires.
* Add generic remote secondmate transport * Add routed remote secondmate replies * Add remote outbox backlog handoff * Integrate remote secondmate lifecycle * no-mistakes(review): Fix remote snapshot and handoff races * no-mistakes(review): Serialize remote home provisioning transactions * no-mistakes(review): Harden remote lifecycle transaction boundaries * no-mistakes(review): Serialize remote lifecycle mutations and fail closed * no-mistakes(review): Close remote lifecycle and file race windows * no-mistakes(review): Serialize remote reply retirement and inheritance * no-mistakes(review): Harden remote transfer integrity and recovery * no-mistakes(review): Serialize remote respawn with registry retirement * no-mistakes(document): Document remote bootstrap convergence accurately * no-mistakes(document): Clarify skipped remote secondmate mutations * no-mistakes(lint): Resolve remote script ShellCheck warnings * no-mistakes: apply CI fixes
* feat(spawn): propagate a native W3C traceparent to spawned agents Add a default-off capability that resolves one W3C traceparent for a task, injects it into the agent's pane shell as the TRACEPARENT environment variable immediately before launch, and records the identical value as traceparent= in state/<id>.meta, so an external observer that explicitly reads that env value or meta field can correlate a worker, a Secondmate, and their nested children into one trace with no collector, storage, UI, or vendor coupling. TRACEPARENT as an environment variable is a firstmate convention carrying a W3C-formatted value: W3C Trace Context standardizes the header, not an env var, and OpenTelemetry SDKs do not read it automatically, so a downstream must consume it deliberately; this feature parents no SDK span by itself. Identity is per task, not per spawn: the carrier is minted with random ids on the first spawn, adopted as a child (fresh span, same trace) for a nested spawn whose parent already holds one, and reused verbatim from the meta on relaunch, so a task keeps one stable logical identity across restarts. A malformed or all-zero inherited value is treated as absent and roots a fresh trace. A new root is sampled (01) - a sampling decision a downstream parent-based sampler honors, not a guarantee that any collector stores a span, and firstmate emits no spans; a child preserves the inherited flag. Trust boundary: a firstmate-minted root is random and reads no prompt, path, task prose, credential, or arbitrary environment key. An inherited TRACEPARENT is opaque caller-controlled data - up to 24 bytes of id passed through after syntax validation - so whoever set it controls those bytes, a bounded fixed-width channel rather than a general content or secret channel. The feature adds no OTEL_* variable, no tracestate, and no arbitrary environment injection; it runs no configurable or arbitrary command, only the fixed local od and tr (resolved from PATH) to read a few bytes of entropy - a small local pipeline with no network or watchdog and no hard latency guarantee. Any entropy or validation failure that returns omits the carrier without aborting the spawn. A default-off spawn leaves the generated meta and launch environment unchanged. Enablement is default-off (config/trace-context, or FM_TRACE_CONTEXT where a non-empty value overrides and unset or empty defers to the file) and is propagated into secondmate homes, taking effect at each agent's next launch: a Secondmate launched or relaunched after enablement carries the primary trace into its nested workers, while an already-running Secondmate roots new traces for its own workers until relaunched. Injection reuses the existing GOTMPDIR channel, so all spawn backends and harnesses and the ship, scout, and secondmate paths are covered. Covered by a pure-library suite and a spawn-path integration test (fake tmux plus a real worktree, hermetic against ambient FM_TRACE_CONTEXT) proving the recorded and injected carriers are identical and sent before launch, that default-off writes and injects neither, that a relaunch reuses the recorded carrier, and that an explicit FM_TRACE_CONTEXT overrides the file both ways; plus a source-owner inheritance test proving trace-context propagates and absence-mirrors through propagate_inheritable_config. Documentation follows the repository documentation-audiences contract: docs/trace-context.md is maintainer-architecture rationale, the configuration schema lives in docs/configuration.md, and the repeatable test evidence is separated into docs/verification/trace-context.md (maintainer-verification), registered in docs/documentation-audiences.json. * fix(spawn): propagate the effective trace-context decision to secondmates FM_TRACE_CONTEXT overrode trace context only in the process that read it. A newly launched secondmate decided enablement from the inherited config/trace-context file alone, so the override did not cross the primary-to-secondmate boundary: FM_TRACE_CONTEXT=off with the file present left the secondmate's nested workers traced (a broken kill switch), and FM_TRACE_CONTEXT=on with the file absent left them untraced despite the inherited carrier. Deliver the primary's effective decision to a newly launched secondmate as a normalized on/off FM_TRACE_CONTEXT in the launch prefix, so a FM_TRACE_CONTEXT override governs the nested primary -> secondmate -> worker chain both ways, not just the copied file. The value is bounded to the literal on/off and does not broaden environment injection; the already-running secondmate boundary is unchanged. Add a genuine two-level spawn regression that drives fm-spawn twice with the exact environment the primary injects into the secondmate and proves both divergent directions end to end. Correct the documentation that implied secondmate coverage on every backend, since orca and cmux reject secondmate spawns, and refresh the verification evidence for the new assertion count. * no-mistakes(review): Clarify Secondmate trace-context launch snapshots * no-mistakes(document): Correct trace-context documentation ownership and relaunch semantics * fix(spawn): resolve the trace-context decision once for carrier and snapshot The effective trace-context decision was read twice per spawn: once inside fm_trace_context_resolve for the recorded carrier, and again for the secondmate FM_TRACE_CONTEXT launch snapshot. A config-file change between the two reads could pair a carrier with the opposite enable state - an injected carrier with an off snapshot, or no carrier with an on snapshot. Freeze the effective on/off decision once, drive the carrier resolution under that frozen FM_TRACE_CONTEXT so it cannot independently re-read the file, and reuse the same frozen decision for the secondmate launch snapshot. Add a spawn-path regression that drives the file-decided path and proves the recorded carrier and the delivered snapshot always agree, and refresh the verification evidence for the new assertion count. * no-mistakes(review): Preserve legacy Secondmate trace boundary * no-mistakes(document): Correct trace-context verification comparison base * no-mistakes(review): Captain, prevent failed trace delivery metadata claims * no-mistakes(review): Captain, align trace-context tests and verification evidence * no-mistakes(document): Correct trace-context verification evidence * no-mistakes(lint): Suppress intentional ShellCheck literal-dollar warnings * no-mistakes(review): Captain: freeze trace context at session start * no-mistakes(test): Captain: stabilize scheduler test and document Kimi trace coverage * no-mistakes(document): Document trace-context safety boundaries * fix(trace): fail off on stale session snapshots Publish each home session decision atomically through a same-directory temporary file and bind it to the current session lock. A replacement failure can no longer leave an earlier on decision active in a later session; missing, stale, malformed, or unpublishable state defaults safely to off. Add regressions for read-only replacement and failed publication, update spawn and session-start fixtures for the lock-bound format, and refresh the architecture and verification records. * no-mistakes(review): Fix trace spawn failure independence and duplicate safety * no-mistakes(document): Refresh trace-context documentation and verification * no-mistakes(review): Clear partial backend input after failed trace submission * no-mistakes(review): Stop unsafe trace delivery before launch append * no-mistakes(document): Document unsafe trace delivery handling * fix(trace): bound each trace to one routed task, never the routing agent A persistent Secondmate holds its launch-time TRACEPARENT in the process environment for its whole life, and routed requests never replace it, so resolving new-task carriers from the ambient environment chained every routed task into one ever-growing trace per Secondmate with distinct parent ids. Resolve now reuses the task's recorded carrier or mints a fresh sampled root, never reading ambient TRACEPARENT, so each routed task is its own trace boundary while relaunch, recovery, and scout-to-ship promotion keep one stable per-task identity. The spawn regression models the reviewed scenario exactly: two unrelated tasks spawned sequentially through one persistent Secondmate environment record and inject distinct trace ids, adopt nothing from the Secondmate's carrier, and a relaunch of the first task reuses its original carrier verbatim. * docs(trace): define the per-task trace boundary The design contract is one task per trace: a persistent Secondmate is routing infrastructure with its own agent identity, never a shared trace root for the unrelated tasks routed through it. Root/recovery semantics replace the removed child-inheritance path, the sampling and safety sections drop inherited-carrier language because ambient TRACEPARENT is never read, and the verification page records the refreshed suite inventories including the two-task Secondmate boundary regression. * test(trace): adopt the explicit per-task delivery contract in spawn fixtures Rebasing onto current main brings the explicit per-task delivery contract: ship spawns now require --mode and --yolo instead of resolving them from the project registry. The trace spawn fixtures pass the same explicit contract canonical spawn tests use, preserving the per-task trace boundary coverage unchanged, and the verification page records the refreshed comparison base.
* fix(bin): classify tmux agent liveness independent of process titles
`fm_backend_tmux_agent_state` attributed a pane solely from
`#{pane_current_command}`, which is a process TITLE a harness can rewrite,
not a structural fact. Claude Code 2.1.220 reports its version string there,
so a live Claude endpoint classified `ambiguous`: the session-start secondmate
liveness sweep could no longer see it, and any consumer that gates on a
positive classification refuses outright.
Read a second, independent name source: the kernel `comm` of every process in
the pane tty's foreground process group. Either source naming a verified
harness yields `alive`, because a false `dead` is the one verdict that can
start a duplicate agent on a live worktree. Scoping to the foreground process
group rather than the pane's descendants keeps a harness-named background
process from faking an agent, and covers multi-process launchers (the Pi
Launcher path) without a special case.
Verified on 2026-08-03 against all seven adapters running for real on tmux
3.6a / macOS 26.5.2 arm64: claude 2.1.220, codex-cli 0.146.0, opencode
1.18.11, pi 0.82.0, pi-signed 0.82.0, grok 0.2.118, kimi 0.31.1 all classify
`alive`, each attributed by a source independent of its title.
Two tests, because they fail for different reasons:
- tests/fm-tmux-agent-liveness.test.sh pins the logic with real processes and
no harness, so it runs everywhere CI runs tmux. It drives the two name
sources apart on purpose and asserts the divergence, so no case can go
quietly vacuous.
- tests/fm-harness-liveness-drift-live-e2e.test.sh relaunches every installed
harness and fails naming the harness and version when one stops being
attributed by a title-independent source.
AGENTS.md section 4 carries the resulting standing rule, and
firstmate-coding-guidelines owns how to satisfy it.
* no-mistakes: apply CI fixes
* docs: move the harness-dependent-check policy out of AGENTS.md
The standing rule was stated in AGENTS.md section 4 with the mechanics in
firstmate-coding-guidelines, which split one contract across two owners and
charged every session for a rule that only fires when firstmate's own
harness-dependent code is being changed.
firstmate-coding-guidelines is now the single owner of both the rule and how
to satisfy it: real-harness proof required, that proof authorized to spend
tokens, structural signals preferred over vendor-rendered surfaces, and a
guard that fails loudly naming the harness and version where a surface signal
is unavoidable. No inline stub is left behind, because AGENTS.md already
carries the load trigger for that skill in sections 7 and 13, so it is read
before any change to firstmate's shared tracked material.
Also records the cross-platform lesson the pipeline caught in the portable
regression, and corrects that file's header: the divergence assertion lives
on the version-string case, which diverges on both supported platforms,
rather than on every case.
* no-mistakes(review): Harden tmux liveness identity and drift validation
* no-mistakes(document): Clarify cross-platform tmux liveness documentation
…#1609) * feat(bin): trace remote secondmate routes and unify the inherit allowlist Per-task W3C trace context (kunchenguid#995) resolved and injected its carrier only at the local spawn path. A remote secondmate is routed through spawn_remote_secondmate, which returns long before that site and wrote its own metadata block, so a remote secondmate stayed silently untraced even with the capability enabled. The parent home still owns that task's identity, because it holds the metadata an observer reads. It now resolves the carrier against the task's own meta under its own frozen decision - reused verbatim on relaunch, freshly rooted otherwise, never adopting the parent process's ambient TRACEPARENT - and hands it to the configured host through a new fm-spawn --traceparent argument, accepted only for a secondmate launch and only as a strict W3C value. The remote host exports it at the same unconditional pre-launch site and reports back the carrier its endpoint actually holds, which the parent records, so an already-alive endpoint reports the identity its agent really received rather than one the parent merely intended. Disabled remains byte-identical and off. The remote inherit path also carried its own hardcoded copy of the inheritable config set, already drifted from FM_INHERITABLE_CONFIG by trace-context. Both remote ends now derive from that one declaration, so a future item cannot be sent by one side and refused by the other, and session-scoped enablement items are skipped on live convergence exactly as the local path skips them. Also fixes a latent stderr leak: an absent session lock printed a raw redirect failure, which the new remote resolve site made visible. Adds tests/fm-remote-secondmate-trace-context.test.sh, driving the real parent -> fm-on -> remote entrypoint -> control -> remote fm-spawn chain over the deterministic SSH boundary and reading the carrier back from the remote pane's own log. * no-mistakes(document): Clarify remote trace and allowlist contracts
* feat(bin): widen the remote runtime PATH and add a remote doctor preflight The fixed remote entrypoint hard-coded a four-directory PATH, so a remote account whose tools live under nix or a per-user profile could not run basic Firstmate work without a login shell. The entrypoint now composes its child PATH from the code root's bin, the account's ~/.local/bin, the common package-manager directories that actually exist on the host, and the portable system tail, deduplicated and in a fixed order, still under env -i with the same variable allowlist and no shell command string. fm-remote-doctor.sh reports that exact PATH by inheriting it from its own entrypoint launch rather than recomposing it, so the ordering keeps one owner. It is read-only, reports where each required and optional tool resolved, and exits non-zero naming every required tool that did not. Remote seeding runs it as a preflight before anything is created on the host and restores the registry when it fails. * no-mistakes(review): Harden remote git authorization and missing-tool diagnostics * no-mistakes(document): Document remote PATH doctor and safe shims * no-mistakes(lint): Fix ShellCheck findings in remote path tests * no-mistakes(lint): Suppress exported fixture's false-positive ShellCheck warning
* feat(bin): gate remote second mates on herdr readiness A remote second mate now always runs on the Herdr backend, whose server belongs to the host's GUI login session and therefore outlives the SSH connections that supervise it. fm-spawn's remote route forces that backend and the host-local control script refuses any other, so the requirement cannot be dropped from either side. fm-remote-doctor.sh becomes the single owner of what "ready" means. It keeps its PATH and tool reporting from kunchenguid#1623 and adds the Herdr, Aqua LaunchAgent, GUI-session, server-reachability, and entrypoint-symlink checks, tagging each gap fixable: or human: with the exact operator step. --fix closes only the automatable gaps - writing and loading the Aqua-scoped dev.firstmate.herdr launch agent, starting the server where no launch agent applies, and recreating the entrypoint symlink - then re-derives every check from the host, so a human gap is never presented as fixed. It never creates a login session, writes an auto-login password, or touches FileVault. Remote seed, remote spawn, and the startup liveness relaunch all run the same check, repair, re-check sequence through one shared library and fail closed with the doctor's own gap text. Recovery inherits the gate because it respawns through the same route. Tests drive the real doctor against a controlled account fixture with a private HOME, a state-backed launchctl, and a fake herdr, and prove the dangerous actions are never attempted. The remote lifecycle suites gain a stateful Herdr CLI fixture and answer the readiness gate at the SSH boundary, so they never inspect or repair the runner's own account. * no-mistakes(review): Validate launch-agent contract and confirm Herdr startup * no-mistakes(review): Validate loaded launch-agent contract before readiness * no-mistakes(review): Refuse legacy remote backends without altering routes * no-mistakes(review): Clarify conditional remote readiness repair sequence * no-mistakes(review): Repair remote readiness before liveness probing * no-mistakes(review): Preserve unknown seeds and reject legacy liveness * no-mistakes(document): docs: clarify remote Herdr backend ownership
…1659) * Pin remote secondmates to fm-remote * no-mistakes(review): Fail closed on legacy remote Herdr endpoints * no-mistakes(review): Isolate fm-remote launch agent from interactive default * no-mistakes(document): Document shared remote Herdr retirement safety
) * feat: run remote commands through Aqua job worker * no-mistakes(review): Enforce remote job deadlines and safe worker shutdown * no-mistakes(review): Refresh stale workers and harden dependency-free supervision * no-mistakes(review): Harden worker ownership recovery and shutdown quarantine * no-mistakes(review): Fix doctor bootstrap, harness repair, and output draining * no-mistakes(review): Probe doctor tools through authenticated worker bootstrap * no-mistakes(review): Refresh stale workers before doctor tool probes * no-mistakes(review): Recover stopped quarantines and extend job deadlines * no-mistakes(review): Separate queue and execution timeout windows * no-mistakes(review): Supervise Linux worker crashes and bind root identity * no-mistakes(review): Resolve authorized Nix profile bin links * no-mistakes(review): Clarify Nix path resolution documentation * no-mistakes(review): Harden PATH safety and nvm selection * no-mistakes(review): Honor nvm system defaults and refresh doctor digest * no-mistakes(review): Keep workers ready during active jobs * no-mistakes(review): Bound pre-execution validation by job timeout * no-mistakes(document): Clarify remote worker documentation * no-mistakes(lint): Fix remote worker ShellCheck diagnostics * no-mistakes: apply CI fixes * no-mistakes: apply CI fixes * no-mistakes: apply CI fixes
* fix(remote): arm SSH dead-peer detection in fm-on.sh A vanished remote host mid-poll (a reboot, a dropped link) left ssh blocked indefinitely on a half-open TCP connection, because fm-on.sh's ssh invocation had no ServerAliveInterval/ServerAliveCountMax. This wedged the remote-reply ferry: fm-procevent.sh's runner blocked inside the ssh child and never reached its own no-result -> claim-release -> reconcile re-arm self-healing path, which otherwise already handles a nonzero exit with empty output correctly. Recovery required a manual retire and re-arm. Arm ServerAliveInterval=15 and ServerAliveCountMax=3 by default (bounded ~45s detection window), both overridable via FM_SSH_ALIVE_INTERVAL and FM_SSH_ALIVE_COUNT_MAX. This is a transport- level fix in fm-on.sh, so it covers every remote command routed through it, not just the reply ferry. The remote sshd answers keepalive probes independently of whatever the remote command is doing, so a legitimately long-but-alive command (a 55s poll, a clone, the doctor) is never falsely killed - only a truly vanished peer trips it, turning that case into a bounded, detectable ssh failure (exit 255) instead of an indefinite hang. Extends tests/fm-on.test.sh with a behavioral regression asserting a bounded, positive ServerAliveInterval/ServerAliveCountMax on the real ssh argv captured through the FM_SSH_BIN process seam, plus coverage that both are env-overridable. * no-mistakes(document): Document SSH dead-peer detection ownership
* feat(bootstrap): gate stale axi CLIs at the floors firstmate actually uses Add gh-axi 0.1.29 floor so bare --squash PR merges stop failing quietly on older builds. Raise tasks-axi to FM_TASKS_AXI_MIN=0.2.2 (multi-id mv) while keeping feature probes. Keep quota-axi at 0.1.16 after verifying schema 3 and per-model availability already ship there; runway remains optional. * no-mistakes(document): Clarify AXI compatibility documentation ownership
…d#1661) * fix(guard): stop false watcher-down alarm mid-turn under Claude auto-arm bin/fm-guard.sh derived its watcher-health verdict from fm_watcher_healthy, which requires a live watcher process holding the home lock. Under the Claude Stop-hook auto-arm supervision model the watcher is armed at each turn end and exits on its wake, so it runs only between turns. Every guarded command run mid-turn therefore found no live watcher and printed the "WATCHER DOWN - SUPERVISION IS OFF" banner even though supervision was healthy. Because the episode key was derived from the beacon mtime (which the between-turns watcher advances every poll), the full banner re-printed on essentially every command, and the message always blamed a "fresh beacon" that was in fact fresh. Make the pull guard's health check model-aware via a new fm_watcher_supervision_verdict in bin/fm-wake-lib.sh: - Under the auto-arm model a beacon fresh within FM_GUARD_GRACE is healthy even with no live watcher process; only a beacon stale beyond grace (or absent) is a genuine lapse and alarms. - Under every persistent-watcher harness (codex foreground checkpoint, opencode/pi/grok background arm, tmux, unknown) a live identity-matched watcher with a fresh beacon is still required, unchanged. The banner now names the true failing condition, a missing live watcher process versus a genuinely stale beacon, instead of always blaming the beacon, and the once-per-episode dedup keys on that condition rather than the beacon mtime so a genuine lapse announces once and does not re-print each turn. The turn-end guard keeps the strict fm_watcher_healthy check because it fires at the turn boundary, where the auto-arm brings a fresh watcher up and it cooperates with that arm. fm_watcher_healthy itself is unchanged, so the arm layer's start/attach/replace decisions are unaffected. Tests in tests/fm-guard-stale-banner.test.sh cover the auto-arm healthy fresh-beacon-without-a-watcher case, the auto-arm stale-beacon alarm and its stable episode, the true-reason banner wording, and the reason-keyed episode surviving a beacon mtime change; existing persistent-model cases are pinned to that model. * no-mistakes(review): Pin secondmate supervision model to launched harness * no-mistakes(document): Align watcher documentation with model-aware supervision health
* fix(tests): stop fixture-tempdir helper from self-deleting under command substitution fm_test_tmproot is almost always called as `TMP_ROOT=$(fm_test_tmproot prefix)`, which forks a subshell to capture its stdout. The old implementation set its EXIT cleanup trap inside that call, so the trap fired - and deleted the fixture root - the instant the subshell exited, before the real caller's own EXIT trap was ever installed. Every test using the documented call pattern leaked its fixture root on every run; two suites had already independently discovered and worked around this with ad-hoc mktemp calls. Registration now goes through a $$-keyed registry file instead of in-process state, since $$ resolves to the invoking shell's PID even inside the subshell. The real cleanup trap is armed once at source time (always the real caller, never a subshell) for EXIT, INT, and TERM. A best-effort orphan sweep on next source reaps marked fixture roots old enough to be from a killed prior run. Simplifies the two existing ad-hoc workarounds (fm-procevent.test.sh, wake-helpers.sh) back onto the shared helper now that it works correctly. * no-mistakes(review): Preserve live fixtures during orphan reaping * no-mistakes(review): Harden fixture ownership against PID reuse * no-mistakes(review): Secure cleanup registry against path precreation * no-mistakes(review): Make fixture registration transactional * no-mistakes(document): Documentation already matches fixture cleanup behavior * no-mistakes: apply CI fixes * no-mistakes: apply CI fixes
* feat(herdr): default presentation spaces on with an explicit opt-out Herdr's disposable one-task presentation workspace was opt-in through the presence of local config/herdr-presentation-spaces. It is now on by default, and a home opts out by writing "off" into that same file. Values are read with the whole-file whitespace-stripped convention the other scalar config items already use, plus case folding. An absent file, an empty file, and "on" all resolve on; only "off" opts out; an unrecognized value warns and keeps the default rather than failing a spawn over a purely visual setting. The empty file is exactly the historical opt-in form, so every home that had already enabled the projection stays enabled with no migration step, and no previously enabled home can be turned off by the flip. Because absence now means on at both ends, secondmate inheritance needs no item-specific convergence: mirroring an absent primary file converges a secondmate to the same default-on rather than turning its projection off, and only an explicit primary opt-out propagates the opt-out. The gate itself moves into fm_backend_herdr_presentation_enabled in the Herdr adapter so the semantics have one owner that regressions can exercise directly. * no-mistakes(document): Document Herdr default-on presentation safety --------- Co-authored-by: kunchenguid <kun-1@kunchenguid.com>
…henguid#1711) * fix: surface consolidated open decisions on every wake-drain A needs-decision or blocked event buried under later, unrelated status appends was only ever shown via the last-line wake annotation, so a still-open captain decision could go silently missed even though status_open_decisions (fm-classify-lib.sh) already folds the whole status stream correctly and fleet-snapshot/bearings already reuse it. Wire that same fold into bin/fm-wake-drain.sh: a new fleet-wide scan_open_decisions wrapper scans every state/<id>.status, and fm-wake-drain.sh prints a separate, bounded OPEN DECISIONS section on every drain (including the empty-queue fast path), so session-start and every wake-handling turn surface it for free without duplicating the open/resolved fold itself. Heartbeat wakes drain through the same script, so this covers that surface too. Also tighten status_open_decisions' file guard to skip an unreadable status file instead of leaking a bash redirection error, now that a fleet-wide directory scan can reach files a single targeted read would not. * no-mistakes(review): Prevent status symlinks leaking open decisions * fix: drop unbounded perl subprocess from status symlink guard The review step's own symlink-safety auto-fix (O_NOFOLLOW read via a perl subprocess) forked one perl process per status file scanned by the new fleet-wide open-decisions scan, with no cap - inflating fm-wake-drain.sh's total external-read cost from 8 (the existing annotation read_cap) to 18 in the enrichment-caps regression test. The plain [ -L "$f" ] check already rejects any status file that is itself a symlink before any read happens, which is exactly what the new regression test exercises and is the same defense level the sibling scan_captain_relevant_statuses/last_status_line already rely on elsewhere in this file (no O_NOFOLLOW). Drop the subprocess-based nofollow read and keep the cheap builtin guard. * no-mistakes(document): Document actionable fleet-wide open decision drains
…kunchenguid#1710) * fix(bin): abort orphaned no-mistakes runs and reap leaked processes at teardown Teardown could remove a task's worker while its no-mistakes pipeline run was still parked at a gate, leaving an orphaned run holding a fleet slot indefinitely (observed 2026-08-03: runs parked 7h39m and parked at a post-CI approval gate). It could also leave backgrounded/disowned descendant processes rooted under the worktree or tasktmp surviving reparented to init (observed: two `go test` binaries pinning CPU for hours with no live task meta to attribute them to). Add two coupled pre-teardown steps, both scoped to this task's exact branch/head or worktree/tasktmp so they can never touch another task's run or processes: - conclude_task_no_mistakes_run aborts a run parked at a gate via `no-mistakes axi abort`, cd'd into the exact worktree so the daemon resolves the run itself rather than teardown naming a --run id. - reap_task_worktree_processes sweeps for processes whose cwd is under the worktree or tasktmp (via `lsof -a -d cwd`) and TERM/KILLs them. Both run before any worktree return, branch delete, or backend kill, and are idempotent on a retried teardown. The branch+head attribution logic is factored out of bin/fm-crew-state.sh into the new shared bin/fm-nm-run-lib.sh so both scripts use the same ownership contract. * no-mistakes(review): Fail closed on incomplete teardown cleanup * no-mistakes(review): Bind teardown cleanup to verified run and process identities * no-mistakes(review): Require confirmed aborts and convergent identity-safe process reaping * no-mistakes(review): Handle process exits during teardown identity checks * no-mistakes(review): Restore teardown library in hermetic gotmp fixtures * no-mistakes(document): Document teardown run attribution and timeout * no-mistakes(lint): Rename shell variable conflicting with done keyword * no-mistakes: apply CI fixes
kunchenguid#1709) The script installs as a symlink under ~/.local/bin. Taking dirname of the symlink itself (instead of its real target) pointed SCRIPT_DIR at ~/.local/bin, breaking sourcing of the sibling fm-remote-job-lib.sh. Resolve the real path first, preferring python3's os.path.realpath, then realpath, falling back to the raw BASH_SOURCE on hosts with neither.
…guid#1754) * fix(bin): preempt remote reply long-polls for queued short jobs Session start on a home with live remote second mates could stall silently for many minutes: the single serial remote job worker ran each armed fm-remote-delta-read.sh reply poll to its full 55s window while bootstrap's short sync, inherit, state, and route commands sat queued behind it, and non-FIFO queue pickup let re-armed polls keep winning the lane. Measured end to end, a trivial short job took 31s behind one 30s poll window. The worker now preempts a running preemptible job (the read-only, cursor- anchored delta read is the only member of that class) as soon as a non-preemptible job is queued, publishing exit 75 with emptied output - byte-identical to the poll's own elapsed-window-with-no-data result - so the parent runner takes its existing no-result path and the watcher re-arms from the same cursor with nothing lost. The delta read translates SIGTERM into that same exit after removing its staging directory. Sibling polls never preempt each other, so two armed monitors cannot churn. The same measured scenario now completes in 1s. * no-mistakes(document): Clarify remote poll preemption documentation
…#1778) Discord mentions already ride the same pairing-token opt-in, relay poll, and platform-aware reply path as X mentions, but the docs still read as X-only, so a stranger could not self-serve the Discord path. Add the numbered turn-on steps to the X mode configuration reference, pointing at the myfirstmate dashboard for account creation, bot install, and token issuance rather than duplicating operator setup here, and drop the X-only framing from the README bullet, the documentation index, and the architecture overview.
…#1781) * feat(bin): run session start deterministically on hook-capable harnesses Session start relied on a native nudge that only asked the agent to run bin/fm-session-start.sh, and an agent can defer that. Observed 2026-08-01: an /ahoy-first session followed the recap path and did not take the helm until a later request forced it. Claude, Codex, and Pi now RUN the digest in their session-open hook through the new bin/fm-sessionstart-run.sh, so the full ordered digest is in model context before the first turn. That wrapper is the single owner of what a session-open source means: startup and Pi's "new" take the helm, clear and compact re-emit, resume/reload/fork delegate to the nudge, and an unreadable source takes the helm because doing that redundantly is idempotent while skipping it is the bug. Grok and OpenCode keep the nudge as the floor, since neither can carry hook stdout into a model turn. Because the hook now blocks session initialization, fm-session-start.sh bounds itself first. Its steps are not all individually bounded - bootstrap's gh auth probe, tool version probes, the backlog listing and per-task endpoint reads are unbounded - so the whole digest runs as one bounded child (default 120s). Whatever it emitted before the bound survives, and the parent adds a loud STARTUP TRUNCATED banner naming the stage that stalled and every stage that never ran, still exiting 0. --reemit skips only the sweeps startup already reconciled. It still re-verifies lock ownership and still drains queued wakes, which arrived after startup and are the turn's work. fm-bootstrap.sh gains FM_BOOTSTRAP_LOCKED so a re-emit keeps repair ownership instead of deferring to a lock holder that is itself. Also adds bin/fm-timeout-lib.sh as the single owner of bounded execution, replacing three near-identical copies, and gives the ahoy skill a helm check so a nudge-tier harness cannot recap before taking the helm. Verified live on 2026-08-05 against Claude 2.1.222, Codex 0.146.0, and Pi 0.82.0; docs/verification/supervision.md records the per-harness source vocabulary, the two named gaps, and the refresh command. * no-mistakes(review): Harden session-start completion, timeout, and Pi delivery * no-mistakes(review): Harden completion ownership and portable timeout escalation * no-mistakes(review): Normalize watchdog KILL exits without masking command status * no-mistakes(review): Guarantee startup bounds and align harness delivery tiers * no-mistakes(test): Fix Pi session-start live verification fixture * no-mistakes(document): Align session-start documentation with deterministic hooks * no-mistakes(lint): Silence intentional child-shell expansion lint warning * no-mistakes: apply CI fixes * no-mistakes: apply CI fixes
* docs: rename the user-facing product name to Relay The public-mention integration gated by the `.env` pairing token is now called Relay across user-facing prose, covering X and Discord alike instead of implying a single network. Renames the product-name strings only: README, docs, the captain-facing skill descriptions, and the AGENTS.md operating prose, including the `X mode (.env)` and `Optional X mode` headings and every link anchor that pointed at them. AGENTS.md section 14 carries a one-line bridge note so the older name and the unchanged identifier spellings stay discoverable. Internal identifiers are untouched: `FMX_*`, `config/x-mode.env`, `state/x-*`, `bin/fm-x-*`, the `fmx-respond` skill path, `__FM_X_MODE_ENV__`, and `x-mode-error`. Platform references to X and Discord as networks stay as they are, and the bootstrap-diagnostics entry still quotes bootstrap's emitted `FMX: X mode on/off` line verbatim because `bin/` output is out of scope for this pass. * no-mistakes(review): Complete Relay prose rename in maintained docs * no-mistakes: apply CI fixes
* feat(harness): add a verified muse crewmate adapter Muse Code joins the fleet as a crewmate/scout adapter, verified live against Muse Code 0.1.0-R708.1 in an isolated lab. Detection matches the anchored prefix muse-bin*, because the installed launcher execs a version-suffixed binary whose name changes on every auto-update and whose install path carries no muse component to fall back on. The same identity is taught to the tmux liveness classifier, without which a healthy muse pane would have read as a dead endpoint. Busy state folds muse's own durable session event log, bound per task by a sessions-root/worktree sidecar. It is a pull source with no writer, so nothing is armed and no record is ever seeded. The fold is anchored on the full run lifecycle prefix so muse's nested cleanup "terminal" payloads cannot settle an in-flight run, and it is depth-bounded so muse's native sub-agent logs cannot be mistaken for the parent's. The idle half stays gated: an open run proves busy, but a settled log reads unknown until a credentialed multi-step run proves one turn stays inside one run. Two findings corrected the scout report. The exec-only --no-foreign-personal-context flag is rejected by the interactive TUI, so the privacy control that actually reaches a pane worker is MUSE_EXPERIMENTAL_FOREIGN_PERSONAL_CONTEXT_KILL, verified to drop the operator's foreign personal rules while keeping the project's own AGENTS.md. And an unauthenticated muse pane never exits, it waits on a device-code prompt, so credentials are a spawn preflight rather than a screen check. muse is refused for secondmates: it has no primary supervision protocol and its hook dialect rejects the reawakening handlers that protocol needs. Per the captain's decision, auto-update is not pinned, and the credentialed multi-step smoke is deferred with an explicit checklist in docs/verification/muse.md. * no-mistakes(review): Accept Muse dispatch profiles and shared efforts * no-mistakes(review): Bind Muse busy state to current session * no-mistakes(review): Compare Muse workspace bindings literally * no-mistakes(review): Harden Muse worker credentials and live signal verification * no-mistakes(review): Cache Muse session bindings and clarify worker credentials * no-mistakes(review): Clear Muse marker inheritance and normalize interrupt aliases * no-mistakes(review): Verify Muse glyph effective foreground color * no-mistakes(review): Harden Muse XDG paths, session cache, and glyph parsing * no-mistakes(document): Document Muse adapter boundaries
…d#1787) * fix(herdr): floor default-on presentation spaces at Herdr 0.8.0 Default-on presentation projection turns every crewmate teardown into a workspace-emptying removal. The focus-safe removal plan avoids Herdr's focus-stealing explicit close only while the doomed pane's shell can be proved lone, childless, and idle; a persistent child of that shell (gitstatusd, a zsh-async worker, direnv) fails that proof permanently and forces the plain close, which on every release before Herdr 0.8.0 moves the captain's active workspace for ~140ms on each teardown. Gate the unconfigured default behind a Herdr 0.8.0 floor. At or above it, project as before; below it, fall back to the flat per-home layout with one warning per home per detected release naming the version and the upgrade. An explicit "on" - including the historical empty opt-in file - is still honored below the floor, so a deliberate opt-in is never silently downgraded. The floor reads two independent signals from the client's own status, either of which can establish a supported release: the protocol number and the release core of the version string. Measured against the real release binaries, no build lacking both upstream focus fixes reaches protocol 19 and every pre-fix build tops out at 17, so protocol 19 is a safe structural expression of the floor. A release that reports neither signal readably is treated as unsupported rather than guessed at. Also: - Correct the adapter comment claiming the mitigation "stays safe without any version gate". That holds for the pane-death route only; the plain-close fallback is reachable precisely on the releases where it is unsafe. - Stop discarding the projected-close helper's stderr at teardown, so a refused or failed focus restore is visible instead of silent. The close stays non-fatal; the presence gate still decides record removal. - Add Part C to the focus-flash regression: a doomed pane whose shell holds a persistent child, in the geometry where the closing workspace's right neighbour is not the anchor. That is the fallback branch the suite could not structurally reach. On 0.7.5 it observes a bounded four-sample wrong-focus window restored exactly; on 0.8.0 it observes none. It also cross-checks its own measurement against the floor classifier, so a drifted protocol mapping fails loudly. - Make the projection suite's unconfigured-home case release-aware, so the whole real-Herdr lane passes on both the CI-pinned 0.7.4 and 0.8.0. - Add an opt-in live guard that re-measures the release-to-protocol mapping against the pinned upstream binaries. The immediate no-code mitigation for a home that cannot upgrade remains writing "off" into config/herdr-presentation-spaces. * no-mistakes(review): Pin Herdr live-guard digests across supported platforms * no-mistakes(review): Document authorized Herdr cleanup containment * no-mistakes(review): Harden Herdr warning marker publication * no-mistakes(review): Honor running Herdr server presentation floor * no-mistakes(review): Recheck Herdr floor after server ensure * no-mistakes(review): Refresh 0.7.5 and 0.8.0 focus transcripts * no-mistakes(review): Route Herdr floor probe through lab session * no-mistakes(document): Align Herdr floor documentation and comments * no-mistakes(lint): Document Herdr presentation out-parameter consumer
* fix(muse): trust the settled session log as idle The credentialed multi-step smoke on Muse Code 0.1.0-R708.1 answered the one question the idle half was held back for: one real 75-second tool-loop turn with 23 tool batches stays inside exactly one run started/terminal pair, and an Escape mid tool loop closes that run as cancelled rather than leaving the turn to continue in another run. A settled log is therefore a finished turn, not a pause between the runs of one turn. Remove fm_busy_muse_idle_verified and FM_BUSY_MUSE_IDLE_VERIFIED_VERSIONS outright rather than pinning them to a version: the session log's own metadata carries only semver 0.1.0 and a build sha, so a version allowlist could not actually match the running build and would be false precision. A settled log now classifies idle, an open run still classifies busy, and only a resolution failure - no binding, no matching log, an unreadable or run-free log - stays unknown. Record the evidence in docs/verification/muse.md, including the run-scoped grep the counts must use, and keep the post-upgrade re-check guidance. * no-mistakes(review): Document Muse idle trust and remove stale gate reference * no-mistakes(document): Clarify Muse idle verification ownership
…guid#1798) * feat(session-start): order the startup digest for truncation safety and bound its bulk The digest is delivered through a harness that truncates an oversized payload from the tail, and it really has been truncated: a 70KB digest arrived as lines 1-435 of 578, cutting off eight lines before the live-task inventory. That session took the helm without ever seeing which tasks were live or where their endpoints were. Three changes, one file's worth of composition: - FLEET STATE is emitted before CONTEXT, so a truncated tail drops curated memory - stable session to session, already governed by a captain-set budget, recoverable with one targeted read - instead of live fleet identity. The LOCK/BOOTSTRAP/WAKE-QUEUE safety preamble keeps its order. The read-once contract moves out of the closing reminder into its own section ahead of both, and now names the condition that voids it: a stage the truncation banner reports as never emitted. - Status-tail lines are capped per line, reusing the cut the wake digest's OPEN DECISIONS section already applies. An observed tail line ran 865 characters and nothing bounded it. The cut and its marker now live in one place, bin/fm-line-cap-lib.sh, so the two digests cannot drift apart; each task's full status log path is still printed beside its tail. - The backlog listing is composed as a recovery input: done rows are never listed, every in-flight, held, and blocked row is shown in full with its hold and blocked-by metadata, and only the dispatchable-now listing is bounded - with an exact remainder count and the command that shows the rest. FM_SESSION_START_QUEUED_LIMIT (default 20) replaces FM_SESSION_START_BACKLOG_LIMIT, which bounded the whole listing indiscriminately and so could drop a held or blocked row. Tests exercise the real digest output: section ordering with the preamble pinned, the per-line cap and its marker, and the backlog composition including the remainder counters on both the tasks-axi and manual paths. * no-mistakes(document): Clarify digest source recovery comments
…kunchenguid#1842) A captain decision opened by a keyed needs-decision:/blocked: status line orphaned as permanently open whenever the answer kicked off work: the worker's next event is working [key=<workstream>] in a different key namespace, so no resolved [key=<decision>] ever landed and the OPEN DECISIONS fold kept listing the answered decision forever. Remove the writer-dependency at its source: the answering firstmate already holds the decision key when it sends the answer, so fm-send's new --resolve-key flag (repeatable) appends the closing resolved line to this home's own state/<id>.status after the submit is confirmed. The close is a local ledger append for crewmates, local secondmates, and remote secondmates alike - a remote mate's escalations reach this ledger through the parent-replies ingest, so only the answer message crosses the transport. Safety: each named key must currently be open per the authoritative status_open_decisions fold or fm-send refuses before sending; a failed or unconfirmed send never closes a key; an append failure after a delivered answer exits nonzero with the manual close command so the decision re-surfaces instead of silently vanishing; a send without the flag closes nothing, and working:/done: still never clear a captain decision. Complementary fixes: the wake-drain OPEN DECISIONS section prints the answer-with-close command hint at the moment of use; brief scaffolds separate resolved's two duties (keyed-phase end vs decision closure) and state that a done:/working: line never closes a decision even when the answer started that work, keeping worker self-close for blockers that clear without a firstmate reply; AGENTS.md and docs/architecture.md carry the one-line pointers to the fm-send contract.
…1836) * feat(secondmate): seed a remote home from a supplied project origin Remote seeding required a local projects/<name> clone purely to read `git remote get-url origin` into the provisioning manifest, so setting up a remote second mate forced disposable clones and no-mistakes inits in the primary home for projects that home has no reason to hold. Firstmate now resolves the origin itself and names it as <project>=<origin-url>. The seed validates and transports what it is given, and the receiving host re-validates it rather than trusting the sender; bin/fm-project-origin-lib.sh is the single owner of which URLs are accepted, refusing executable remote-helper transports, option-shaped values, and unusable spellings at both ends. A bare <project> still reads an already-present clone's origin, so nothing that works today has to change. Registry consistency is unchanged: an unregistered or local-only project is still refused. A remote seed therefore creates nothing in the primary home beyond the route, the charter, and its launch record. The lifecycle test now seeds a registered project the primary has never cloned and asserts the primary project tree is byte-identical afterwards, alongside refusals for a missing origin, an unsafe origin, a local-only project, and an unregistered project. * no-mistakes(review): Clarify project origin documentation ownership * no-mistakes(document): Document supplied-origin remote seeding contract * feat(secondmate): accept project origins from any host or forge Firstmate is a shared template, so a project origin must be able to name any host: GitHub Enterprise on a private domain, GitLab hosted or self-hosted, Bitbucket, Gitea, Codeberg, sr.ht, a bare IP, an SSH config alias, or a plain server nobody else has heard of. The validator already decided on structure rather than on a forge allowlist, and this makes that guarantee explicit and closes the two gaps that a host-agnostic rule exposed: - a bracketed IPv6 literal in the scp-like form is now accepted, so a host reachable only by address is not excluded - a "/../" traversal inside a local or file: origin is now refused, because that names a path on the cloning host's own filesystem The library is the single owner of the accepted forms, and its header says plainly that there is no host, domain, or forge allowlist and there must never be one. The skill keeps its distinct agent-operating lines (the agent resolves and supplies the origin; a remote seed creates nothing in the primary home beyond the route, the charter, and its launch record) and points at the library for URL acceptance and at the operator doc for the rest. The lifecycle test now drives Bitbucket, a self-hosted enterprise domain, a self-hosted GitLab over ssh with a port, and a bare scp-like custom host through the real seed, manifest, transport, and remote provisioning path in one seed, asserting each URL reaches git unchanged and each clone carries its own origin's content. The unit matrix leads with non-GitHub hosts for the same reason. * no-mistakes(review): Validate project origin authorities safely * no-mistakes: apply CI fixes * no-mistakes: apply CI fixes
…guid#1851) * fix(tests): copy the whole bin/ tree into the old-vs-new conformance shim main went red on tests/fm-backend.test.sh's "fm-send --key: old vs new exit code" assertion, which reads as an fm-send fail-closed regression from build_old_bin enumerated by hand the sibling scripts it copied into the synthetic pre-refactor tree. kunchenguid#1842 made bin/fm-send.sh source bin/fm-line-cap-lib.sh (added by kunchenguid#1798) and the list never learned about it, so the pinned old fm-send.sh aborted at `. "$SCRIPT_DIR/fm-line-cap-lib.sh"` under set -eu and exited 1 before parsing a single argument, while the current one delivered the key and exited 0. The parity check compared a crashed process against a working one and reported a behavior divergence that never happened - the more so because BASE_REF collapses to HEAD on main, where both sides run byte-identical source and a genuine divergence is impossible. fm-send's --key exit path is unchanged and its fail-closed contract is intact. Copy the tree whole instead of enumerating it. An enumerated list has to be extended by hand every time an entrypoint gains a dependency and is the only thing that knows; it has been patched a dozen times for exactly that. A whole-tree copy has nothing to forget. Extracting a refactored entrypoint the baseline does not have now fails loudly instead of writing an empty file. Only old-vs-new parity covered that exit contract, and parity is near-vacuous on main. Pin it directly: tests/fm-send-strict.test.sh drives delivery both ways from one stub and asserts an undelivered key exits nonzero naming the key, so swallowing that error fails the suite. * no-mistakes(review): Materialize historical fixture dependencies from baseline * no-mistakes(document): Clarify fm-send key regression scope
* fix(bin): mirror the whole remote secondmate status stream A remote secondmate's reply channel required corr=<16hex> on every line and failed the entire delta when one line lacked it, so the cursor could never advance past that line and the channel wedged permanently. The charter tells a secondmate to report its own progress phases and to raise new decisions with no correlation token, because correlation only answers a marked parent request. Those lines were therefore unrepresentable on the remote channel, while a local secondmate writes them straight into the parent's status file. Treat the channel as what it is: a mirror of the mate's status stream. A remote mate now presents the same status and decision model as a local one, so a newly raised needs-decision reaches the parent's open-decision fold identically, and correlation goes back to being a per-line property that settles a pending request rather than a gate on the stream. Only what crossing a machine boundary genuinely adds stays behind: cursor continuity, confined document fetch and rewrite, at-most-once append, and control-byte normalization that rewrites bytes without ever dropping a line. Line framing and size bounding already belong to fm-remote-delta-read.sh. A document the remote reader refuses is named in one escalation instead of stalling the stream, while an unavailable transport still leaves the delta for the existing retry. * refactor(bin): give the remote reply stream one append owner Every line entering the parent status stream - a mirrored line, the continuity escalation, and the undelivered-document escalation - now goes through one at-most-once append, so the idempotence a replayed generation depends on is stated once instead of copied at three call sites. * no-mistakes(review): Keep local document transfer failures retryable * no-mistakes(review): Isolate reply headers and normalize payload bytes * no-mistakes(review): Correct remote reply mirror contract wording * no-mistakes(review): Update remote reply script catalog description * no-mistakes(document): Document remote status-stream mirroring
…chenguid#1847) fm_secondmate_parent_record_parse read the .fm-secondmate-parent record with bash's read, which drops NUL bytes - and different bash generations disagree on the result: 3.2 truncates the value at the NUL while 5.x splices the surrounding bytes together. A NUL-bearing parent_home could therefore resolve to a home the record's bytes never name contiguously, and which home fm-teardown.sh's promised-public-reply resolution read (registration, registry, relay state) - or whether that protection engaged at all - depended on which interpreter ran the cleanup. Reproduced end to end: the same NUL-bearing record cleaned up under bash 5.x by resolving the spliced-together registered parent, while bash 3.2 refused it as unresolved, and a literal truncated path refused under both. Reject any NUL byte in the record before field parsing, putting corrupt records in the same fail-closed bucket as duplicate fields, malformed local bindings, unsupported routes, and symlinked records. The regression test drives the real bin/fm-teardown.sh over the proven clean-cleanup fixture with a NUL spliced mid-path into the recorded parent_home, so before the fix it reproduced the wrong-home cleanup and now it must refuse with the explicit binding refusal.
…unchenguid#1853) * docs(secondmate-provisioning): require record intake for an inherited domain A new mate seeded for an existing or inherited domain previously pulled in charter, inherited config, captain-shared preferences, project clones, and queued backlog rows with zero instruction about the domain's shipped history, so it assumed a greenfield domain. A live backlog keeps only the configured recent Done entries, so an inherited queue structurally over-represents plans and under-represents deliveries, and already-delivered work resurfaced as open. Add a record-intake step to the creation/seed path: classify greenfield versus existing or inherited, and for the latter reconcile every inherited plan against origin/main plus the live deployment, take only genuinely open work and still-live durable knowledge, never carry a plan row for shipped work, and record what could not be reconciled. Greenfield domains are untouched. The skill owns the procedure; the backlog handoff section carries a one-line reinforcement at the point where plan rows actually move. * no-mistakes(document): Clarify secondmate record-intake scope
…guid#1860) * perf(session-start): run every network check off the blocking path The session-start digest runs on a session-open hook that blocks session initialization, and every external-network call it made was individually unbounded: `gh auth status`, secondmate liveness, secondmate convergence, pending remote handoff delivery, and the fleet-sync fetch. One unreachable remote secondmate could consume the whole FM_SESSION_START_TIMEOUT and truncate the digest, so a slow network could cost the work queue itself. Measured against a host hanging 25s per SSH connection, that startup took 1m18s. The digest is now composed from local reads alone. bin/fm-startup-network.sh runs the same checks concurrently in a bounded detached worker and the digest harvests whatever finished, without ever waiting. Same fixture: 0.84s. Nothing is dropped. fm-bootstrap.sh stays the single owner of every sweep and still runs all of them, through a new FM_BOOTSTRAP_NETWORK phase split whose `skip` and `only` halves are a partition of the unsplit run. Deferral is safe because the sweeps are idempotent detectors, the result is durable and always surfaces (inline, or as a `check: startup-network` wake), and the worker re-verifies that the fleet lock still names the session that asked before it mutates anything. While the worker is still running the digest names exactly what is unconfirmed rather than implying it passed. A relaunch performed by the deferred pass is now always reported, because the digest that printed the superseded endpoint record is already out. Also collapses the duplicate tasks-axi compatibility probe: the verdict is computed once and handed to the bootstrap child for one process hop, then consumed so it never reaches a spawned agent's environment. 10 tasks-axi invocations per startup become 7. Verified on Claude Code 2.1.222 that a worker detached by the session-open hook survives the hook returning, the one vendor behavior this design needs and no portable test can see. Re-landed on current main, superseding PR kunchenguid#1845, which was cut from a pre-kunchenguid#1842 base. The digest's section numbering in AGENTS.md section 3 now states the emission order directly - supervision block and its read-once contract, fleet state, network checks, then context - which keeps kunchenguid#1826's fleet-state-before-context ordering. The old-bin test shim keeps main's git-archive baseline from kunchenguid#1851, which already subsumes this branch's reason for widening that shim. * docs(verification): re-measure the deferred startup stage on the current base Re-runs the unreachable-remote latency fixture against default-branch tip 8398d31 rather than the now-historical 345de4e, and records the sweep-result comparison the deferral's safety argument rests on: the deferred worker's published report is byte-identical to the three sweep lines the blocking baseline printed, with the unreachable route preserved in both. * no-mistakes(review): Fail deferred startup when report publication fails * no-mistakes(document): Document deferred startup network behavior accurately
* fix(procevent): apply a captured adapter result in code, not by instruction A remote secondmate's reply was captured and announced, but never applied. Nothing dispatched the reply adapter's `handle` on a `procevent remote-reply` wake, and the handling instruction named only the generic acknowledgement, so the wake was retired while everything it carried was dropped: the reply never reached the secondmate's local status mirror, the request it answered kept escalating as a missed report, and the relay - whose registration each capture retires, and which only that same handling re-arms - was left dead until the next session start armed it again. Applying such a result carries no judgement, so it belongs in code. After publishing, the runner now calls `bin/fm-procevent-<adapter>.sh autohandle <source-id> <sequence> <result-file>` and lets the adapter apply and acknowledge its own result, through the same kind of seam that already owns the terminal verdict. It runs strictly after terminal retirement, because a handling adapter re-arms its own next source and retiring afterwards would drop that fresh registration. An adapter with no such command, or one whose pass does not complete, leaves the result unacknowledged and therefore still announced, so a handler receives it exactly as before. Resolving the request was not enough on its own either. An escalation opens a durable keyed decision in the parent status log, and nothing ever closed it, so a request the remote had answered kept surfacing in every later open-decisions fold. The pending-reply library now owns both ends of that decision: it opens one under a per-request key rather than the shared default key, and closes it once the record resolves, appending the closing line only while that exact decision is still open in the fold so it can neither double-close nor clear an unrelated decision that has since taken the same key. The handling instruction still routes a wake to its adapter, now as the idempotent confirmation of what the runner already did rather than as the guarantee. Verified end to end in a throwaway isolated home driving the real armed source, blocking delta reader, runner, and wake queue, with the handler doing only the generic acknowledgement and no part of the ingest stubbed: before, seven failed observations reproducing the incident; after, none. Each half is independently load-bearing - without the runner change the reply never reaches the mirror, without the escalation close the settled request still surfaces as an open decision. * no-mistakes(review): Prevent legacy reply closure from masking decisions * no-mistakes(review): Serialize pending reply resolution and escalation closure * no-mistakes(review): Serialize pending reply escalation with resolution * no-mistakes(review): Clarify guarded legacy escalation closure behavior * no-mistakes(review): Guard legacy closure and reserve pending reply keys * no-mistakes(review): Match pending reply escalations by construction * no-mistakes(document): Document automatic remote reply resolution * no-mistakes(lint): Fix unused concurrent escalation loop variable * no-mistakes(lint): Fix unused concurrent resolution loop binding * no-mistakes(review): Version fold cache and gate autohandle on publication * no-mistakes(document): Clarify remote reply relay documentation
…kill (kunchenguid#1841) Bring the public installer-facing stow skill up to the internal skill's current curation behavior while keeping it fully standalone: - Replace the total-capture thesis with the compact-operating-map framing. - Add read-the-destination-before-writing with the inspect-then-update triad (supersedes what, one-sentence rewrite, delete stale now). - Add the concrete prune list together with its unique-fact guard, as an accuracy discipline with no size-budget machinery. - Curate every memory file the pass has open, not only the routed one. - Add the standing-decisions sweep category. - Add the stronger-owner pointer-over-copy test before filing. - Add tool-agnostic task-note discipline (inspect, classify, considered replacement body, never blind-append) and blocked-on recording. - Give .stow-notes.md a closed set of three exits. - Forbid storing, creating, or editing a skill as a stow destination. - Report per-file action verbs in the completion receipt. - Consolidate the repeated local-vs-external and .gitignore prose and fix the second-person voice slip, so the file does not grow (11334 -> 11276 bytes).
…nchenguid#1917) * fix(hooks): keep tracked Claude entries inert under grok 1.0.0 hooks Grok loads Claude-compatible settings, so the tracked `.claude/settings.json` hook entries also fire under Grok. They were meant to be inert there, guarded by `[ -z "${GROK_AGENT:-}" ] || exit 0`. That guard silently stopped working. Verified from the live process environment of a wedged grok 1.0.0 Stop hook on 2026-08-07: a grok 1.0.0 HOOK process carries GROK_HOOK_EVENT, GROK_HOOK_NAME, GROK_SESSION_ID, and GROK_WORKSPACE_ROOT, but no GROK_AGENT. The observed hook process was labelled `GROK_HOOK_NAME=project/settings:stop[0].hooks[1]`, which is the Claude-only auto-arm entry. Consequence: Grok ran `bin/fm-claude-stop-autoarm.sh` synchronously. Grok has no `asyncRewake`, so it waited on the foregrounded watcher for that entry's declared 28800-second timeout and the Grok turn never ended - the operator saw an infinite "Responding". Widen the guard to `[ -z "${GROK_AGENT:-}${GROK_HOOK_EVENT:-}" ] || exit 0` on the five entries that have a `.grok/hooks/` counterpart: both Stop entries, the SessionStart entry, and the two PreToolUse Bash entries. Two deliberate limits: - The guard is NOT widened to GROK_SESSION_ID. Grok injects it into every child process, so it can survive into a Claude session that Grok launched and would silently disable Claude's own watcher continuity. GROK_HOOK_EVENT is per-hook-invocation and does not leak that way. - `bin/fm-subagent-pretool-check.sh` stays unguarded on purpose. It is the one tracked entry with no `.grok/hooks/` counterpart, so guarding it would remove the guard from Grok entirely rather than deduplicate it. The new test asserts it stays unguarded so the exception cannot be closed silently, and docs/subagent-guard.md is honest that the coverage it leaves is partial. `bin/fm-harness.sh` corrects a comment that presented GROK_AGENT as reliably present; it is a fast path only, and the ancestry walk is what actually guarantees grok identification. tests/fm-turnend-guard.test.sh adds test_tracked_claude_entries_inert_under_grok, which runs every tracked entry under a real grok 1.0.0 hook environment, a legacy GROK_AGENT environment, and a native Claude environment. * no-mistakes(document): docs: sync grok hook-marker guard facts to owners * no-mistakes(review): docs: state grok guard criterion by event coverage
… stage (kunchenguid#1918) The deferred network stage published one aggregate started/finished pair, so a run that took a minute could not be attributed to a phase, a host, or a clone without re-running it by hand under manual tracing. Add bin/fm-timing-lib.sh as the single owner of elapsed-time records, and bracket each network owner with one: the gh auth probe, the secondmate liveness sweep, secondmate convergence, pending handoff delivery, and the project clone refresh, plus one record per secondmate for the remote-touching steps (id and host) and one per project clone. Each record carries a start offset from one shared origin, so the artifact reads as a timeline. The stage publishes them beside its report as state/.startup-network.timings, for a timed-out or failed run too, where the partial record is the answer. Only the on-demand `report` command prints them: `harvest` composes the session-start digest, so its output, the wake cadence, and every other part of a normal session start are unchanged. Recording is inert unless a run asks for it, so nothing else that sources these scripts pays for it. Details are identities only - a detail carrying whitespace is refused rather than cleaned up, which is what keeps a command line, an environment dump, or a captured error out of the file. Split two per-item loop bodies into their own functions so each iteration can be timed; every `continue` became a `return 0` with the same meaning, and the sweeps still run directly, in the same order, returning the same results.
…kunchenguid#1928) * feat(stow): cascade the internal /stow to every registered secondmate Invoked in a primary home, /stow now sweeps every registered secondmate after the primary's own required pass, enforcing the same startup-memory threshold in each home against that home's own allowance rather than a fleet total. bin/fm-stow-cascade.sh owns the mechanical inputs: it enumerates each registered secondmate exactly once from data/secondmates.md, reports that home's own budget accounting, and resolves how the sweep reaches it. A live agent sweeps its own home so its uncaptured session knowledge is captured too; a local home without one is curated in place; a remote home without one is accounted read-only and deferred, because there is no generic remote write path for a home's own memory files. Every host- crossing step and each home's accounting runs under one hard bound, so a slow or unreachable home reports an exception and the sweep continues. Nothing changes until /stow is invoked: no new notification, digest section, or background work. The public skills/stow skill is untouched. * no-mistakes(review): fix(stow): extend cascade --help range to include full exit-code contract
…nguid#1927) 29 fm-remote-job-worker.sh processes were found running at ppid 1, 1-2 days old, each still polling and appending to a log inside a no-mistakes gate worktree that had already been returned. Three things combined to make that possible: - The recorded worker.pid is the serving child, not the restart supervisor above it, so a teardown that stops that one pid only makes the supervisor respawn. The Linux start path also left the worker tree in the launching command's process group, so there was no group to signal instead. - Neither the serving loop nor the supervisor ever rechecked whether its configured FM_ROOT still existed, so a worker launched from a worktree outlived that worktree indefinitely. - The supervisor restarted a failing child with a fixed 0.1s delay and no bound, which is what grew the logs (~66MB/day measured). The Linux start path now puts the worker tree in its own process group, and fm_remote_job_stop_worker_tree signals that whole group - refusing any group whose leader is not itself a worker, so a worker from an older build or from launchd's own session is still stopped safely as a single process. The worker stops itself once its code root stops being a Firstmate checkout, confirmed across a grace window so an ordinary transient cannot stop a healthy worker. The supervisor backs off and gives up rather than restarting forever. bin/fm-remote-job-reap-orphans.sh is the belt-and-suspenders sweep for workers already orphaned that way, wired into fm-teardown.sh. Its reap condition is exactly "the code root named in the worker's own command line is gone", which is why the account's healthy LaunchAgent worker and every live remote secondmate worker are never candidates. The two suites that leaked these in the first place now stop the worker tree rather than the recorded pid alone.
…henguid#1925) * fix(bin): lint only the changed shard locally, full lint in CI Two ships hitting fm-lint.sh at once could spike CPU to 190% and load to 8.58 on a captain's Mac, even though each run finishes quickly. fm-lint.sh now defaults to linting only the canonical-set files changed since the merge-base with origin/main (including uncommitted edits) on an ordinary local branch, using plain local git with no network calls. It still lints the full canonical set in CI (GITHUB_ACTIONS=true or CI=true), on the main branch, or whenever no merge-base can be found, so CI coverage never depends on a local diff. Explicit paths keep bypassing this selection entirely. * no-mistakes: apply CI fixes
* feat(bin): add deterministic agent lifecycle control Separate firstmate's data plane from its control plane. bin/fm-send.sh is the data plane: conversational text, always routing-marked for a kind=secondmate target. That marking is right for a message and wrong for a lifecycle command - a marked "/quit" arrives as ordinary chat the agent reasons about instead of executing. bin/fm-control.sh is the control plane: allowlisted interrupt, exit, and transactional relaunch verbs addressed to an exact task id, with per-harness mechanics owned by the executable bin/fm-control-lib.sh rather than improvised in agent prose, and a verified postcondition for every action. There is no arbitrary-text and no raw-key entry point. relaunch runs as a transaction with a durable journal: it resolves the profile, proves the work it must preserve is recoverable, records the required progress note, stops the old agent, then delegates the launch to its single owner, bin/fm-spawn.sh --relaunch, which adopts the recorded endpoint and worktree instead of creating either. A refusal before the stop leaves the record and instructions byte-identical; a failure after it reports the concrete state rather than claiming an agent that is not running. Teardown and discard stay separate and explicit. exit and relaunch require a backend with a recovery-grade agent-state classifier, so zellij, orca, and cmux are refused rather than reported as successful blind. A remotely placed secondmate is refused by name, because its agent runs on a host where none of these postconditions can be read. * fix(control): resolve a recorded harness to its adapter before retiring wiring fm-spawn arms per-task harness wiring on prefixes, because a task launched from a raw command records that command's basename rather than the exact adapter name. The control plane's retirement tables are keyed by the exact adapter, so a task recorded as `grok-2` had its turn-end token, private registry entry, and worktree hook pointer armed and never retired - leaving a registry entry that outlived the agent that owned it. State the prefix rule once, in the capability owner, and resolve the recorded value through it before every table lookup. bin/fm-send.sh's composer-clear lookup reads the same owner instead of keeping its own copy of which adapters need one. * test(control): pin muse session-binding retirement across a harness switch * no-mistakes(review): Resolve prefixed harnesses across lifecycle control verbs * no-mistakes(review): Report interrupt delivery without fabricating cancellation state * no-mistakes(review): Clear disabled relaunch trace context atomically * no-mistakes(review): Clarify control interrupts and restore legacy send state * no-mistakes(review): Refuse ambiguous relaunches and report exit delivery * no-mistakes(review): Revalidate interrupts and accept interrupt-stopped exits * no-mistakes(review): Lock descendant tasks before forced recursive teardown * no-mistakes(document): Align lifecycle adapter documentation with control plane * no-mistakes: apply CI fixes * fix(bin): serialize fresh task publication with forced teardown Forced secondmate teardown enumerated a home's task set, locked what it found, then re-enumerated while removing. A fresh spawn takes only its own per-task lock, so a record published inside that window was invisible to the preflight and visible to the cleanup: it was destructively processed while never lifecycle-locked. Reproduced with real agents. A record published 0.249s after teardown began was removed, its window closed, and its worktree returned to the pool - while both commands reported success. A per-task lock cannot protect a task that does not exist yet. Add a per-home task-set lock guarding WHICH tasks a home has, as opposed to the metadata lock guarding one task's record. Teardown takes it per home, parent before child, before enumerating and holds it through cleanup. A fresh spawn takes it before its own per-task locks and holds it through publication; a relaunch is exempt, because it republishes an existing task already covered by that task's control lock. Either the spawn publishes first and the teardown's preflight covers it, or the teardown owns the set and the spawn refuses. Both directions fail closed, and both are pinned by tests that hold the lock rather than racing on timing. * no-mistakes(review): Serialize remote secondmate publication with forced teardown * no-mistakes(review): Preserve remote spawn routing and state initialization * no-mistakes(review): Serialize teardown when descendant state is absent * no-mistakes(review): Cover symlinked descendant state refusal * no-mistakes(document): Document task-set serialization safeguards * no-mistakes(lint): Isolate task-set lock path resolution * no-mistakes: apply CI fixes
…n before final summary
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Intent
Require firstmate's brief scaffold to declare which local skills a crewmate must read for ship and scout briefs, so established local tooling is never silently skipped
What Changed
bin/fm-brief.shnow hard-gates ship and scout brief scaffolding on a local-skill declaration: the caller must setFM_LOCAL_SKILLS(a single path or a comma/newline-separated list) or pass--no-local-skills; omitting both fails the scaffold before any brief is written, and a value that reduces to nothing after trimming/splitting is treated as unset.FM_LOCAL_SKILLSis set, the generated brief injects a "Local skills - read first" section listing each declared skill/resource and instructing the crewmate not to write a new one-off script unless the skill genuinely doesn't cover the task.AGENTS.mddocuments the new contract, andtests/fm-brief.test.sh(plus call-site updates intests/fm-ask-user-authority.test.sh,tests/fm-secondmate-safety.test.sh, andtests/fm-tangle-guard.test.sh) cover the gate, the injected section content, the mutual exclusivity of--no-local-skillswithFM_LOCAL_SKILLS, and blank-value handling.Risk Assessment
✅ Low: The actual scope of this change (commits 2f83d57, 71cba77, b5d484c touching bin/fm-brief.sh and tests/fm-brief.test.sh) cleanly implements a hard gate requiring FM_LOCAL_SKILLS or --no-local-skills on ship/scout briefs, correctly excludes secondmate charters, updates AGENTS.md so the calling orchestrator knows about the new required input, and updates every other call site across the test suite so no unrelated test regresses. The round-1 finding (whitespace/separator-only FM_LOCAL_SKILLS silently bypassing the gate) was fixed by validating the post-split bullet list instead of the raw env var, and a real regression test (test_local_skills_gate, executing the actual script and asserting exit code plus file absence for FM_LOCAL_SKILLS=',,' and ' ') was added, satisfying the test-quality bar. No injection risk from interpolating FM_LOCAL_SKILLS into the unquoted heredoc (bash performs a single, non-recursive expansion pass). The rest of the base..head diff (mode/--yolo/resolved-line logic, docs) belongs to unrelated already-merged history reachable only because the given base commit predates several unrelated merged PRs; it is not part of this change and was excluded from review.
Testing
Ran the targeted fm-brief.test.sh suite (all 20 cases pass, including the new gate test and blank-FM_LOCAL_SKILLS regression), then manually drove bin/fm-brief.sh as an end user would across the four behavior branches (no declaration → hard fail for both ship and scout; --no-local-skills → success without the section; FM_LOCAL_SKILLS → success with the section correctly rendered), confirming the CLI transcript matches the stated intent. No issues found; manual verification scratch directory was removed and the working tree is clean.
Evidence: Manual CLI transcript exercising all four gate branches (missing declaration, --no-local-skills, FM_LOCAL_SKILLS)
Pipeline
Updates from git push no-mistakes
✅ **intent** - passed
✅ No issues found.
✅ **Rebase** - passed
✅ No issues found.
🔧 **Review** - 1 issue found → auto-fixed ✅
tests/fm-brief.test.sh:195- The follow-up fix in 71cba77 correctly closes a real gap in bin/fm-brief.sh: previously[ -n "${FM_LOCAL_SKILLS:-}" ]treated a value likeFM_LOCAL_SKILLS=",,"orFM_LOCAL_SKILLS=" "as present, so the hard gate was skipped and a '# Local skills - read first' section was written with zero actual bullets (silently producing a broken/empty declaration instead of failing loudly, undermining the stated goal that local tooling is never silently skipped). The fix now validates the post-trim/post-split bullet list instead of the raw env var, which is correct. However, no regression test was added for this exact case (FM_LOCAL_SKILLS set to a value that reduces to nothing after comma-splitting and trimming) in tests/fm-brief.test.sh's test_local_skills_gate — the existing test only covers 'unset' vs 'set to real values', so this bug class isn't guarded against reintroduction.🔧 Fix: test: add regression coverage for blank FM_LOCAL_SKILLS values
✅ Re-checked - no issues remain.
✅ **Test** - passed
✅ No issues found.
bash tests/fm-brief.test.sh (20/20 ok, exit 0)FM_HOME=... bin/fm-brief.sh demo-ship-1 my-proj --mode no-mistakes (no flags) → exit 1, no brief.md writtenFM_HOME=... bin/fm-brief.sh demo-scout-1 my-proj --scout (no flags) → exit 1, no brief.md writtenFM_HOME=... bin/fm-brief.sh demo-ship-2 my-proj --mode no-mistakes --no-local-skills → exit 0, brief.md written, no Local-skills sectionFM_HOME=... FM_LOCAL_SKILLS='skills/seo-audit/SKILL.md, scripts/deploy-check.sh' bin/fm-brief.sh demo-scout-2 my-proj --scout → exit 0, brief.md written with correctly rendered Local skills sectionAGENTS.md:494- AGENTS.md:494 ("explicitly require firstmate-coding-guidelines before editing") and CONTRIBUTING.md:62 / firstmate-coding-guidelines/SKILL.md:72-73 (manual load-by-hand for firstmate-repo briefs) describe a pre-existing, separate mechanism that now sits adjacent to the new FM_LOCAL_SKILLS declaration gate. They are not contradictory and predate this change, but a future consolidation could route the firstmate-coding-guidelines requirement through FM_LOCAL_SKILLS instead of the by-hand instruction. Left untouched as an out-of-scope design decision, not a doc staleness bug.✅ **Lint** - passed
✅ No issues found.
✅ **Push** - passed
✅ No issues found.