fix: avoid false stalls for supervising secondmates

[bingb0t5]

Intent

Stop Firstmate from treating an idle manager pane as stalled when that manager has busy workers. Close the hole at classification time. A manager with at least one busy or working child must not be classified as a stalled/wedge pane solely because the manager's own pane is idle. An idle secondmate with no unread instruction and with live child work must not emit stale:/possible-wedge escalation or secondmate wake-loop stalled checks. Keep true wedges: looping, unresponsive, or unacknowledged instruction. Do not absorb blocked push-transitions or mask failed/blocked/needs-decision/captain-held because a child is busy. Captain-held must survive masking status lines. Push to bingb0t5/firstmate fork only.

What Changed

• Treat quiet local secondmate managers as working when their registered home contains active child work, preventing false stale and possible-wedge wakes.
• Suppress aged secondmate wake-loop checks while child work remains active, with bounded caching and immediate rechecks for unread instructions.
• Preserve real escalations for blocked or failed states, open decisions, captain-held transfers, declared waits, and actionable push transitions.

Risk Assessment

🚨 High: The central child-work inheritance path contradicts the explicit requirement that failed manager state survive masking status lines, so the durable fix remains incomplete.

Testing

Exercised the real watcher, stale-pane classification, and secondmate wake-loop paths. Idle managers with active children stayed quiet, while unread instructions, failed/blocked states, open decisions, and captain-held states still escalated. All targeted commands passed. CLI transcript evidence was captured; no screenshot was applicable because this change has no rendered UI surface.

Evidence: Idle manager with busy child E2E evidence

Source: Idle manager with busy child E2E evidence

Firstmate idle-manager classification - end-to-end evidence
Target: 587e85e1765e2116ff59f2aafe6ab4546a3f6c0f

Executed public watcher/supervision test interfaces:
  bash tests/fm-watch-triage.test.sh
  bash tests/fm-wake-queue.test.sh
  bash tests/fm-secondmate-safety.test.sh

Observed product behavior:
  PASS: idle kind=secondmate manager with busy child crew is not stale
  PASS: idle manager whose own log says working is absorbed while its child crew runs
  PASS: secondmate wake-loop stall stays quiet when manager supervises active child work

Retained true escalation behavior:
  PASS: unanswered captain hold still re-surfaces while the manager's children run
  PASS: secondmate wake-loop stall still fires when the manager has unread instructions
  PASS: secondmate wake-loop stall still fires when the manager holds an unanswered decision
  PASS: secondmate wake-loop stall still fires when a later line masks a captain hold
  PASS: classifier does not let active child work mask manager failed, parked, blocked, paused, open-decision, or captain-held state

All three commands exited 0.

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

⚠️ **Review** - 1 error

• 🚨 bin/fm-classify-lib.sh:1352 - Intent requires that a busy child must not mask failed, but the durable quietness guard only folds open decisions and captain-held transfers. Concrete path: failed [key=deploy]: rollout broke followed by working: dispatched recovery, with an idle manager and a working child. fm-crew-state returns working · status-log from the last line, status_open_decisions ignores failed, and this function reports quiet, so the child changes the classification to working and suppresses escalation. Track unresolved failures at this shared status-fold boundary, or otherwise prove they were explicitly cleared before allowing inheritance.

✅ **Test** - passed

✅ No issues found.

• bash tests/fm-watch-triage.test.sh
• bash tests/fm-wake-queue.test.sh
• bash tests/fm-secondmate-safety.test.sh
• git status --short confirmed testing left no worktree changes

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

_{Reviews (4): Last reviewed commit: "no-mistakes: apply CI fixes" | Re-trigger Greptile}

[kunchenguid]

Speaking as Kun's firstmate: first look on current main 6a2cd6cd2c64 (#3024). Never messaged the captain.

class=corrective. Stops false stall/wedge on an idle kind=secondmate manager whose registered home still has working children. Inheritance applies only after the manager's own verdict is unknown or working · status-log, and only when manager_owes_captain_nothing (durable folds for unresolved failed, open decisions, captain-held, plus last-line captain-relevant). Failed/blocked/paused/parked/done and unread instructions still escalate. Wake-loop skip is a bounded per-row supervising cache, not a permanent mute. Not a new default product.

VISION.md (inspected bin/fm-classify-lib.sh crew_absorb_class / status_unresolved_failures / status_held_decisions / manager_inherits_child_work, bin/fm-watch.sh secondmate_wake_stall_tick supervising cache, bin/fm-wake-lib.sh marker write, bin/fm-captain-hold.sh settle_answered_status_holds, tests/fm-watch-triage.test.sh / tests/fm-wake-queue.test.sh / tests/fm-secondmate-safety.test.sh):

• One captain, one interface: aligns (false stall is noise; true holds and failures still surface).
• Authority is explicit: aligns (no new grant; captain-held cannot rot behind a busy child).
• Scripts own the mechanics: aligns (status-fold + child working class are exact).
• A restart is a non-event: aligns (markers/receipts/cache on disk, bounded by stall cadence).
• Delegation with a spine: aligns (supervision stays honest; workers are not trusted as a mask).
• The fleet outlives any vendor: aligns (local secondmate home only; remote mates excluded).
• Scope: aligns.

This HEAD: 3f166d68043f742b5285f0d92238e71d658f2f3d. MERGEABLE / UNSTABLE, ahead 11 / behind 0.
Attestation 587e85e1… ≠ THIS HEAD. The published review error (busy child masking failed) is addressed on THIS DIFF by status_unresolved_failures inside manager_owes_captain_nothing; the body attestation is still the older SHA.
CI: fork Actions on THIS HEAD are action_required (CI 32831861454, Require no-mistakes 32831861596) — not green. Greptile 5/5 is not a merge gate. Author body does not use a GitHub closing link.

Not first-time. No workflow approval this pass.

Security: none. Local status folds and home ownership file; supervising marker is a regular state file with the same path guard as the stall marker.

Overlap: classify-lib / watch / wake-lib with open #2801 and wedge #3001; same-author sibling #3031. docs/architecture.md / docs/configuration.md shared with other opens. Not lock / spawn-freshen / herdr / #2804.

Land-eligible rec: NO (attestation mismatch; CI not green). Captain-flag NOW: no.

This is waiting-on-author for a HEAD-matching no-mistakes-pipeline-attestation:v1 (and green Actions). Not a captain-decision. Not a merge I will recommend.