fix(bin): only tell no-mistakes workers to pause for compaction when idle-compact is enabled

[NewAiCoder]

Intent

Every new no-mistakes worker stops after its implementation commit with "paused: awaiting compaction before validation" and waits for an idle-compact hand-off that never comes, because this home has idle-compact turned off (only config/idle-compact.off exists, no config/idle-compact). Firstmate has had to steer each worker by hand to start validation. The fix: the Definition of done in bin/fm-dod-lib.sh should only tell a worker to pause for compaction when the home actually has idle-compact enabled; otherwise it should tell the worker to start validation right after its commit. Never run gh issue close, gh issue reopen, or any gh project command.

What Changed

• bin/fm-dod-lib.sh: added fm_dod_compaction_step, which uses the idle-compact watcher's own fm_idle_compact_threshold_minutes check on config/idle-compact. The no-mistakes Definition of done now includes the "paused: awaiting compaction before validation" instruction only when idle-compact is enabled. When it is off, the brief tells the worker to start validation (no-mistakes axi run) right after its implementation commit.
• docs/configuration.md: the declared-state fast path now says the pause instruction applies only when idle-compact is enabled, and describes the idle-compact-off behavior.
• tests/fm-brief.test.sh: added coverage for the brief's Definition of done (+38 lines).

Risk Assessment

✅ Low: The DoD now derives the compaction pause from the watcher's own fm_idle_compact_threshold_minutes verdict with identical config resolution, falls back safely to "start validation" on any failure, and the tests exercise the real fm-brief.sh output for absent, empty, 0 and off configs, so both prior findings are verifiably fixed.

Testing

I ran the repo's fm-brief test file, which passes including the new regression test; that was a unit test run, not a live product drive. I also generated real no-mistakes briefs with bin/fm-brief.sh in a disposable home across seven config states, plus the base commit as a before check. With only idle-compact.off, the brief says to start validation, and the base commit still emits the pause line. Enabled, empty, and commented-numeric files keep the pause. The off, 0, and absent cases drop it, and a config override resolves correctly from another cwd. The temp home and worktree were removed and the working tree is clean.

• Live validation: ✅ go - 5 of 6 scenarios driven live against the product

Scenario	Result	Live	Evidence
Home with only config/idle-compact.off (the reported setup): generated no-mistakes brief tells the worker to start validation and has no compaction pause	✅ pass	live	Live fm-brief.sh output: pause=0 start=1
Regression reproduced: base commit 5b9f42c in the same home still emits 'paused: awaiting compaction before validation'	✅ pass	live	Base worktree brief: pause=1
Idle-compact enabled (empty file or comment plus 20): brief keeps the compaction pause and omits the start-validation line	✅ pass	live	empty-enabled and comment-20 both pause=1 start=0
Adversarial: config/idle-compact present but disabled for the watcher ('off' or '0') or removed: brief starts validation, not the pause	✅ pass	live	off-text, zero and removed all pause=0 start=1
FM_CONFIG_OVERRIDE is honored and the brief resolves it when run from a different cwd	✅ pass	live	override case pause=1 with the override dir enabled
Repo regression test suite for fm-brief, including the new idle-compact config test	⏸️ untested	no	The prior payload recorded this as a unit test run (tests/fm-brief.test.sh) with live=false, so it did not establish a live result against the running product. Drive the generated brief live (covered…

Evidence: Live brief output per config state

only-off (config/idle-compact.off only): pause=0 start=1
empty config/idle-compact: pause=1 start=0
comment + 20: pause=1 start=0
'off' text: pause=0 start=1
'0': pause=0 start=1
file removed: pause=0 start=1
FM_CONFIG_OVERRIDE with idle-compact=10, other cwd: pause=1
Base commit 5b9f42c with only idle-compact.off: pause=1 (the reported bug)
Generated line with idle-compact off: Right after that implementation commit lands, start validation: measure the lane below, then run `no-mistakes axi run`.

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

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

• ⚠️ bin/fm-dod-lib.sh:408 - The new gate treats idle-compact as enabled whenever the file config/idle-compact exists ([ -f "$config/idle-compact" ]). The comment calls this "the same presence test" as bin/fm-idle-compact.sh, but the watcher does not decide enablement on presence alone. fm_idle_compact_threshold_minutes (bin/fm-idle-compact.sh:187), called first in fm_idle_compact_tick (:918), returns disabled when the first content line is not a positive integer. Failing sequence: config/idle-compact contains 0, off, or other non-numeric text. The watcher stays inert and never hands off. The DoD still emits "paused: awaiting compaction before validation". The worker stops and waits forever, which is the stall the change set out to fix. Fix: derive the DoD gate from the same predicate, either by sourcing and calling fm_idle_compact_threshold_minutes "$config" >/dev/null or by sharing one helper. Add a test with a 0 or off file expecting the start-validation text.
• ℹ️ docs/configuration.md:604 - docs/configuration.md (the Declared-state fast path paragraph) still says a no-mistakes ship brief tells the worker to append the pause line unconditionally. Behavior is now conditional on config/idle-compact being enabled, so the doc is stale.

🔧 Fix applied.
✅ Re-checked - no issues remain.

✅ **Test** - passed

✅ No issues found.

• Live validation: ✅ go - 5 of 6 scenarios driven live against the product

Scenario	Result	Live	Evidence
Home with only config/idle-compact.off (the reported setup): generated no-mistakes brief tells the worker to start validation and has no compaction pause	✅ pass	live	Live fm-brief.sh output: pause=0 start=1
Regression reproduced: base commit 5b9f42c in the same home still emits 'paused: awaiting compaction before validation'	✅ pass	live	Base worktree brief: pause=1
Idle-compact enabled (empty file or comment plus 20): brief keeps the compaction pause and omits the start-validation line	✅ pass	live	empty-enabled and comment-20 both pause=1 start=0
Adversarial: config/idle-compact present but disabled for the watcher ('off' or '0') or removed: brief starts validation, not the pause	✅ pass	live	off-text, zero and removed all pause=0 start=1
FM_CONFIG_OVERRIDE is honored and the brief resolves it when run from a different cwd	✅ pass	live	override case pause=1 with the override dir enabled
Repo regression test suite for fm-brief, including the new idle-compact config test	⏸️ untested	no	The prior payload recorded this as a unit test run (tests/fm-brief.test.sh) with live=false, so it did not establish a live result against the running product. Drive the generated brief live (covered…

• bash tests/fm-brief.test.sh (includes new test_no_mistakes_dod_compaction_pause_follows_idle_compact_config; all ok)
• Live bin/fm-brief.sh <id> some-proj --mode no-mistakes in a disposable FM_HOME with only config/idle-compact.off
• Same command with config/idle-compact empty, containing a comment plus 20, containing off, containing 0, and removed
• Same command with FM_CONFIG_OVERRIDE pointing at a dir with idle-compact=10, run from a different cwd
• Same command from a throwaway worktree of base commit 5b9f42c with only idle-compact.off, to reproduce the original bug

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.