fix(watcher): make check wakes lossless via watcher-side suppression

[e-jung]

What Changed

• Moved check-wake suppression from opaque check scripts into the watcher: checks now print their current state idempotently, and the watcher dedups against .seen-check-<name> and enqueues to the durable queue before advancing that marker, so a crashed watcher or lost stdout can no longer swallow a check wake.
• Added a catch-all backstop that scans .babysit-*.seen sidecars and force-escalates a swallowed terminal (MERGED/CLOSED) transition within one sweep, deduped via .escalated-<sidecar>, as a safety net for legacy edge-triggered checks.
• Hardened fm-fleet-snapshot.sh: bounded secondmate-home validation with run_timed (a hung NFS/automount mount can no longer block the parent snapshot) and switched the seen-homes dedup from space-substring to newline-exact matching.

Risk Assessment

✅ Low: The incremental fix commit correctly resolves both prior warnings — validate_secondmate_home is now bounded by run_timed (with correct globals-via-stdout and rc 124/non-zero/zero handling, safe under set -u, no injection) and the seen_homes dedup now uses exact newline-delimited line matching (verified empirically: distinct homes and the space-prefix case are added, only genuine duplicates are caught) — and it introduces no new issues; the only remaining branch item is the acknowledged no-op watcher catch-all tradeoff from the prior round, requiring no action.

Testing

Completed 1 recorded test check.

• Outcome: ⚠️ 1 error across 1 run (19m21s)

Pipeline

Updates from git push no-mistakes

⏭️ **intent** - skipped

✅ No issues found.

⚠️ **Rebase** - 1 warning

• ⚠️ .agents/skills/bearings/SKILL.md - branch carries 1 commit(s) that exist on your local main branch but were never pushed to origin/main; rebasing would bundle this unrelated work (11 file(s)) into the PR:
• 8934c17 fix: derive bearings from authoritative secondmate state (fix: derive bearings from authoritative secondmate state #555)

Push main to origin, or rebase your branch onto origin/main, before gating.

🔧 **Review** - 3 issues found → auto-fixed ✅

• ⚠️ bin/fm-fleet-snapshot.sh:1016 - validate_secondmate_home is called directly (line 1016) without a run_timed wrapper, unlike every other cross-home read in this same commit (registry line ~430, parent-activity ~517, terminal ~546, summary ~1027). validate_secondmate_home (fm-ff-lib.sh:122-189) does ~8 stat/cd+pwd/cat syscalls against the home and its data/state/config/projects subdirs. A registered secondmate home on a stale NFS/automount path or hung mount blocks the parent snapshot for the mount timeout (60s+) with no recourse, directly violating the file's stated invariant at lines 62-63 ('one broken or unexpectedly large home cannot hang the parent snapshot'). Fix: route through run_timed and treat 124 as a validation failure; note validate_secondmate_home sets globals (VALIDATED_HOME/VALIDATION_ERROR) the caller reads, so the wrap must capture the result rather than run in a bare subshell.
• ⚠️ bin/fm-fleet-snapshot.sh:1020 - The seen_homes dedup uses a space-delimited substring match: case " $seen_homes " in *" $home "*). If one validated secondmate home's canonical path is a space-prefixed prefix of another's (e.g. /a/foo registered before /a/foo bar, verified empirically: ' /a/foo ' matches inside ' /a/foo bar '), the second home is wrongly flagged 'invalid home: duplicate resolved home route' and its structured state is silently dropped from the snapshot. Unlikely for canonical secondmate paths but legal, and the fix is mechanical: use a newline-delimited set with exact per-line equality instead of substring matching.
• ℹ️ bin/fm-watch.sh:685 - The new catch-all backstop (lines 685-700) dedups only against its own .escalated-<sidecar> marker, which is decoupled from the regular check path's .seen-check-<name> marker. For a check that both prints a terminal state idempotently AND carries a .babysit-*.seen sidecar (a hybrid/transitional check, or a migrated check with a leftover stale sidecar), the regular path wakes on sweep N and the catch-all emits one additional 'check: catch-all' wake on sweep N+1 for the same transition; the two records carry different wake keys (script path vs sidecar path) so drain does not collapse them. Bounded to once per terminal value (.escalated is then set), self-healing, and consistent with the stated lossless/belt-and-suspenders intent, but worth knowing since either path can change independently.

🔧 Fix: bound secondmate home validation and dedup exactly
✅ Re-checked - no issues remain.

⚠️ **Test** - 1 error

• 🚨 tests failed with exit code 1
• command -v tmux >/dev/null || { echo "tmux is required for e2e tests" >&2; exit 1; }; tmux -V; rc=0; for t in tests/*.test.sh; do echo "== $t =="; bash "$t" || rc=1; done; exit "$rc"

⚠️ **Document** - 1 info

• ℹ️ AGENTS.md:108 - The state/ inventory line (.hash-* .count-* .stale-* ... .seen-* .hb-surfaced-* .last-* .heartbeat-streak watcher internals; never touch) does not enumerate the new .escalated-* watcher-internal marker this change introduced (written by the catch-all backstop). .seen-check-* is already covered by the existing .seen-* glob, but .escalated-* matches no glob on that line. This is not a falsehood — the line is an illustrative category label that already omits the pre-existing .babysit-*.seen marker, and the change authoritatively documents .escalated-* as a suppression marker in AGENTS.md:557. Adding only .escalated-* (without .babysit-*.seen) would make the inventory inconsistent; both-or-neither is the consistent choice, and .babysit-*.seen is out of scope for this change. Left as-is.

🔧 **Lint** - 1 issue found → auto-fixed ✅

• ⚠️ linter found issues (exit code 1)

🔧 Fix: suppress SC2016 on intentional bash -c eval string
✅ Re-checked - no issues remain.

✅ **Push** - passed

✅ No issues found.

[kunchenguid]

Thanks for the PR. Wheelhouse found that this branch currently has a merge conflict with the base branch, so it is stepping out of the maintainer queue for now.

Please rebase on or merge the base branch into your branch, resolve the conflict, and push the result. Once GitHub reports the PR as mergeable again, Wheelhouse will resurface it for maintainer review.

_{Conflict noted for firstmate#34 at ae2d59d2.}

[kunchenguid]

Automated reminder: thanks for the PR! This branch currently has a merge conflict with the base branch.

When you get a chance, please rebase onto (or merge) the latest base branch, resolve the conflict, and push. After that, checks will re-run and the PR will get looked at again.

_{Noted for firstmate#34 at a56af233.}

[e-jung]

Closing as not planned because I have migrated my own orchestration workflow from FirstMate to Orca and will not keep this conflict-heavy branch current. The branch and discussion remain available as history. Thank you for the project and for considering the contribution.