feat(bearings): fleet snapshot, view, and /bearings report (bearings stack)

[leo1oel]

Ports upstream's bearings stack — a "pick up where I left off" report over the live fleet — adapted to the fork's herdr-native, Claude-only model.

What it adds

• bin/fm-fleet-snapshot.sh — the read-only structured foundation (--json, schema fm-fleet-snapshot.v1): backlog (In flight / Queued / Done), one row per task meta with reconciled current state and herdr endpoint presence, open-decision hints, scout reports, and a secondmate-landed roll-up. No locks, wakes, or mutations.
• bin/fm-fleet-view.sh — a human fleet table rendered from the snapshot JSON.
• bin/fm-bearings-snapshot.sh — a compact TOON projection for the catch-up read; local-only by default (zero network), with --include-prs the sole opt-in that touches GitHub.
• .agents/skills/bearings/SKILL.md — the /bearings skill: a dated data/status-report-<date>.md plus a four-section chat digest (Captain's Call / Recently Landed / Underway / Charted Next).

Fork adaptation

The snapshot is written against upstream's multi-backend abstraction; the fork is herdr-native with no fm_backend_* layer, so its backend integration was rewritten:

• Meta reads via a plain grep helper instead of fm_meta_get.
• Endpoint presence via fm_herdr_pane_exists (and secondmate agent-liveness the same way) instead of fm_backend_target_exists/fm_backend_agent_alive.
• Dropped the fm-backend.sh source — in this fork it is a CLI that dispatches on $@ when sourced, not a library.
• A fork-native open-decision model: the fork emits no [key=...] decision markers, so upstream's keyed status_open_decisions fold would degenerate; replaced with tracking the last unresolved needs-decision/blocked line, cleared by a later resolving verb, and reconciled against fm-crew-state (a provably-resumed run-step clears a stale gate).

fm-fleet-view.sh and fm-bearings-snapshot.sh port verbatim (plain gh is present in the fork env, so the opt-in PR path needs no change).

Validation

Local gate:

• shellcheck bin/*.sh tests/*.sh — clean (exit 0).
• All 25 tests/*.test.sh pass, including the two new tests:
  • fm-fleet-snapshot-view — schema, backlog parsing, a task row's kind/endpoint/current-state, the open-decision hint AND its lifecycle clearing when a run resumes, and the view render.
  • fm-bearings-snapshot — the projected schema, the local-only default (a stubbed gh is never invoked), and the landed roll-up from the Done backlog.

---

Note

Medium Risk
Large new surface area touches supervision hooks (fail-closed arm policy) and fleet state interpretation; read-only snapshot paths are lower risk than the PreToolUse deny paths on every Bash tool call.

Overview
Adds a bearings stack for “pick up where I left off” reads: a read-only fm-fleet-snapshot.sh (fm-fleet-snapshot.v1 JSON over backlog, task metas, reconciled crew state, herdr endpoint presence, open-decision hints, scout reports, and secondmate-landed roll-up), fm-fleet-view.sh as a human table renderer, and fm-bearings-snapshot.sh as a bounded TOON projection (local-only by default; --include-prs is the only network path). The new bearings skill drives /bearings: one dated data/status-report-<date>.md plus a fixed four-section chat digest, read-mostly with no fleet mutations.

Fork-specific snapshot behavior uses grep-based meta reads, fm_herdr_pane_exists for endpoints, and a last-unresolved needs-decision/blocked open-decision model cleared when fm-crew-state shows the crew resumed—instead of upstream’s multi-backend / keyed decision markers.

Also wires Claude PreToolUse Bash hooks in .claude/settings.json: fm-arm-pretool-check.sh (fail-closed on bad watcher arms and broad pkill -f fm-watch) and fm-cd-pretool-check.sh (blocks persistent top-level cd/pushd/popd in the primary checkout only), backed by shared shell lexing in fm-arm-command-policy.mjs and sibling fm-cd-command-policy.mjs, with docs and behavior tests.

^{Reviewed by Cursor Bugbot for commit 2a51552. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_cd6240a7-dce9-4d9e-86dc-7a951481534b)