fix(bin): accept post-colon decision keys

[Bloh09]

Intent

Fix the decision-key grammar mismatch between what workers naturally write and what the key parser accepts.

Observed bug (live on multiple tasks): workers write needs-decision: [key=x] ... (key AFTER the colon). _fm_decision_key in bin/fm-classify-lib.sh parsed that as key default, so fm-send --resolve-key x refused with "no open decision with that key" while wake-drain's OPEN DECISIONS section displayed the raw line as if it were keyed. The canonical grammar the parser expected is needs-decision [key=x]: ... (key BEFORE the colon), but brief scaffold rule 6 mentioned [key=<slug>] without showing placement, so workers guessed the post-colon form.

Fix with one semantic owner, robustness first:

1. Make _fm_decision_key (bin/fm-classify-lib.sh) the single owner of key extraction and have it tolerate BOTH placements: needs-decision [key=x]: ... and needs-decision: [key=x] ... (same for blocked/paused/resolved verbs that carry keys). The first key token found in either position wins; absent key still folds to default. Mid-note [key=...] prose must not become a key. Invalid post-colon key tokens fall through to default rather than dropping the line.
2. Update fm-brief.sh scaffold rule 6 to SHOW the exact canonical grammar with a concrete example (needs-decision [key=my-slug]: summary), so new briefs teach the preferred form.
3. Regression tests: both placements for open and resolved lines, including that a post-colon-keyed open decision is closed by --resolve-key and by a resolved line in either form. The live failure (post-colon key parsed as default -> resolve-key refusal) must be covered by a test that fails on the old parser.
4. OPEN DECISIONS incremental fold and fm-send's --resolve-key path both go through the shared owner so they can never disagree. Bump FM_OPEN_DECISIONS_FOLD_VERSION when fold semantics change.

Constraints: firstmate shared tracked material; shellcheck-clean; behavior tests through public interfaces. Acceptance: both key placements parse to the same key everywhere (fold, display, --resolve-key), scaffold rule 6 shows the canonical grammar with an example, regression tests cover both forms, no-mistakes pipeline green with a PR on origin (Bloh09/firstmate) against main.

Prior run failed only because the claude test agent hit a session limit after review already applied fold-version bump + invalid post-colon defaulting; local targeted suites already pass on the current head.

What Changed

• Accept decision keys both before and immediately after the colon, using shared extraction for decision folding, display, and --resolve-key.
• Fall back to the default key for invalid post-colon tokens and bump the incremental fold version for the new semantics.
• Teach the canonical key-before-colon grammar in generated briefs and add regressions for open, resolved, and --resolve-key flows.

Risk Assessment

✅ Low: Captain, the change is well-bounded and satisfies the required grammar, shared-fold consistency, cursor-version migration, scaffold guidance, and behavioral regression coverage without a substantiated source defect.

Testing

Targeted classifier/fold, fm-send --resolve-key, and brief-generation behavior checks passed; manual evidence generation captured the emitted canonical rule and the end-to-end post-colon resolve regression result, and the worktree remained clean.

Evidence: Generated brief rule 6 showing canonical decision-key grammar

6. If a decision belongs to a human (product choices, destructive actions),
   append `needs-decision [key=my-slug]: {summary of options}` (key token BEFORE the colon) and stop. Firstmate will reply with the decision.
   A decision or blocker you opened stays open until a `resolved` line carrying its exact key lands; a later `done:` or `working:` line never closes it, even when the answer is what started that work.
   Firstmate's reply normally writes that closing line at answer time; when a blocker or wait clears WITHOUT a firstmate reply, append `resolved [key=my-slug]: {how it cleared}` yourself (same key placement if you opened it with one) as you resume.
7. Never stop, restart, or update the shared `no-mistakes` daemon - it is one instance serving

Evidence: Public fm-send post-colon regression result

ok - fm-send --resolve-key: post-colon worker key is closable

ok - fm-send --resolve-key: the answer send itself closes the open decision
ok - fm-send --resolve-key: post-colon worker key is closable
ok - fm-send --resolve-key: an answer that starts a workstream leaves no orphaned decision
ok - fm-send: a send without --resolve-key never closes a decision, and working/done still cannot
ok - fm-send --resolve-key: a key that is not open refuses loudly before anything is sent
ok - fm-send --resolve-key: a failed send never closes the decision
ok - fm-send --resolve-key: one answer closes each named key and only those
ok - fm-send --resolve-key: a marked local-secondmate answer closes with the plain answer text
ok - fm-send --resolve-key: a remote-secondmate answer closes the same local ledger, transport-only difference
ok - fm-send --resolve-key: a failed remote transport never closes the decision
ok - fm-send --resolve-key: --key, empty message, explicit targets, and malformed keys refuse loudly

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

✅ **Review** - passed

✅ No issues found.

✅ **Test** - passed

✅ No issues found.

• bash tests/fm-watch-triage.test.sh — exercised shared key extraction, full/incremental decision folding, both open/resolved placements, invalid post-colon fallback, and mid-note exclusion
• bash tests/fm-send-resolve-key.test.sh — exercised the real fm-send and wake-drain interfaces, including the original post-colon-key resolve failure
• bash tests/fm-brief.test.sh — exercised generated ship, scout, and secondmate briefs
• FM_HOME="$evidence_dir/brief-home" bin/fm-brief.sh evidence-scout alpha --scout followed by extraction of emitted rule 6 — captured the generated user-facing grammar contract

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.