fix(spawn): enforce exact reasoning effort for harness launches

[coreldh]

Intent

Ensure fm-spawn never records a requested reasoning effort unless its selected harness launches with that exact effort, or refuse before metadata and launch. Codex CLI 0.147.0 was empirically verified to accept model_reasoning_effort=max for gpt-5.6-luna, so Codex must emit max rather than silently omit or remap it. Apply the same fail-closed invariant to lower-ceiling and no-effort-flag adapters, without changing config/crew-dispatch.json, and publish only through no-mistakes. Kun retains merge authority.

What Changed

• Enforce exact requested reasoning-effort propagation in fm-spawn, including Codex max, with fail-closed refusal before metadata and launch for unsupported adapter levels.
• Align dispatch validation, harness guidance, and regression fixtures with the fail-closed effort contract while preserving config/crew-dispatch.json unchanged.

Risk Assessment

✅ Low: The authorized fixes now align dispatch validation, spawn behavior, documentation, and Muse regressions with the fail-closed effort invariant; the remaining duplicated effort lists are mechanically consistent and do not create a demonstrated reachable defect.

Testing

All focused suites completed with exit 0, demonstrating Codex max emission and fail-closed unsupported-effort refusal before metadata/launch; bootstrap also produced its terminal validation result.

Evidence: Focused regression evidence

Muse: max refuses before metadata/launch. Dispatch: Codex emits max model_reasoning_effort; unsupported adapters refuse before metadata/launch. Bootstrap validator exits 0.

- Outcome: 🔧 2 issues found → auto-fixed ✅ across 2 runs (25m9s)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

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

• 🚨 bin/fm-spawn.sh:1305 - The new Codex branch accepts and emits max at fm-spawn.sh:1305, but the shared dispatch validator still rejects codex:max at bin/fm-bootstrap.sh:1005. A dispatch-selected Codex max request therefore cannot reach fm-spawn; update the earliest shared validation boundary if dispatch profiles are required to express this newly supported effort. This conflicts with the required criterion that Codex must emit max.
• ⚠️ bin/fm-spawn.sh:1325 - Muse max now correctly refuses instead of mapping to ultra at fm-spawn.sh:1325, but tests/fm-muse-harness.test.sh:255 still requires --reasoning-effort 'ultra' and will fail against the new fail-closed contract. Update that regression to assert refusal before metadata and launch.
• ⚠️ .agents/skills/harness-adapters/SKILL.md:151 - The edited harness guidance still contains the obsolete rule at .agents/skills/harness-adapters/SKILL.md:151 that fm-spawn records unsupported effort in metadata and omits the launch flag. That directly contradicts the required fail-closed invariant and the newly updated adapter table; replace it with the refusal-before-metadata behavior.

🔧 Fix: Fixed dispatch validation and Muse fail-closed regression
✅ Re-checked - no issues remain.

🔧 **Test** - 2 issues found → auto-fixed ✅

• ⚠️ tests/fm-muse-harness.test.sh - Muse fixture could not execute: available bash is Mach-O and exits 137 before detection.
• ⚠️ tests/fm-bootstrap.test.sh - Bootstrap test was interrupted without a terminal summary.
• bash tests/fm-spawn-dispatch-profile.test.sh
• bash tests/fm-muse-harness.test.sh
• bash tests/fm-bootstrap.test.sh (interrupted)

🔧 Fix: Muse fixtures now use executable Perl ancestry and both focused regressions pass
✅ Re-checked - no issues remain.

• bash tests/fm-muse-harness.test.sh
• bash tests/fm-spawn-dispatch-profile.test.sh
• bash tests/fm-bootstrap.test.sh
• Verified config/crew-dispatch.json is unchanged in the target diff.

✅ **Document** - passed

✅ No issues found.

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

• ⚠️ linter found issues (exit code 1)

🔧 Fix: Suppress intentional Perl interpolation ShellCheck warnings
✅ Re-checked - no issues remain.

✅ **Push** - passed

✅ No issues found.