fix(bin): distinguish validation handoff from terminal completion

[connectwithclayton]

What changed

• Generated no-mistakes briefs now use needs-validation: {summary} for the committed implementation handoff and reserve done: PR {url} checks green for terminal completion.
• Centralized status classification treats the new marker as actionable and non-terminal, so supervisors can distinguish the two states by verb alone.
• Supervision keeps the two-phase flow intact: workers still stop after the implementation commit, and firstmate still triggers validation before the terminal PR completion report.

Defect

Worker instructions reused the done: prefix for both "implementation committed, start validation" and "PR open, checks green, finished."
A supervisor reading a status log therefore could not distinguish the opposite lifecycle states mechanically.

Before, both states appeared as done::

• done: implementation committed
• done: PR https://example.test/pr/7 checks green

After, the handoff and terminal state are distinct:

• needs-validation: implementation committed
• done: PR https://example.test/pr/7 checks green

Validation

At shipping head 3db189f8, the test phase reported:

Targeted behavioral suites and an end-to-end generator/classifier transcript completed successfully, proving the worker handoff remains non-terminal, active validation supersedes the handoff, terminal PR completion remains done:, and heartbeat fail-safe behavior remains covered; lint was intentionally not run under the test-phase prohibition.

The outer lint phase subsequently ran bin/fm-lint.sh successfully with pinned ShellCheck 0.11.0 and actionlint 1.7.12.

[greptile-apps]

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains; the previously reported legacy FM_CAPTAIN_RE override path now recognizes the validation handoff before applying the override regex.

_{Reviews (2): Last reviewed commit: "fix(classify): preserve validation hando..." | Re-trigger Greptile}

[connectwithclayton]

Addressed the review finding in 1afa3dc. status_is_captain_relevant now recognizes the configured needs-validation handoff before applying FM_CAPTAIN_RE, so a legacy operator regex cannot silently suppress the validation trigger; regression coverage includes both a realistic legacy regex and an overridden handoff verb. I audited every vocabulary consumer: watcher, daemon, stale and signal scans, and push-transition handling all route through the shared classifier, so no other override-blind path remains. Full local runner on the exact committed tree contents: FM_TEST_SUMMARY total=160 failed=12 skipped_gate=27 duration_ms=6472283; all changed-path tests and lint passed, while the 12 failures were unrelated existing baseline cases.