Noticed while implementing #8117 (PR #8159). Observation class, tooling papercut — not a defect. Unassigned; grading is the triage seat's call.
The observation
scripts/pm/dispatch-gates.mjs carries a --self-test mode — 28 cases after #8159, covering workflow extraction, script resolution, watch-hint scanning, the runnable-invocation rendering and the new change-kind derivation. No CI job runs it. Measured on current main:
grep -rn "dispatch-gates" .github/workflows/*.yml — no hits;package.json has no check:* script pointing at it (the two scripts/pm/ scripts that DO have gates are check:pm-skill-ratchet and check:pm-skill-id-lint).
So the self-test executes only when a human or agent types it. A change that breaks the extraction functions lands green, and the breakage surfaces as a dispatch prompt quietly naming the wrong gate families — which is the failure this tool exists to prevent, and the kind that reads as correct output.
Why it is not simply the same as #8110
#8110 (check:skill-frame-freshness --self-test runs nowhere) is the same class but has a different remedy, which is why this is filed beside it rather than as a comment on it:
check:skill-frame-freshness is a CI gate — its invocation exists and the fix is to extend it to --self-test && …, the conventional script shape used everywhere else in this repo;dispatch-gates.mjs is advisory tooling with no CI job at all — there is no invocation to extend, so wiring it needs a deliberate decision about whether a PM-loop tool earns a gate slot, and if so where it hangs.
Both are worth deciding together; neither answer implies the other.
Directions (not a recommendation — grading is triage's)
- Give it the conventional gate shape — a
check:pm-dispatch-gates script that is --self-test only (the live derivation reads the workflow tree and would be a slow, noisy gate), wired wherever the two existing scripts/pm/ gates run.
- Fold every
scripts/pm/** self-test into one job — one step that runs each pm script's --self-test, which also covers whatever lands there next without a per-script decision.
- Do nothing. The tool is advisory; a wrong gate list costs the PM one dispatch round, and the PM reads real per-job conclusions afterwards anyway.
Related
Noticed while implementing #8117 (PR #8159). Observation class, tooling papercut — not a defect. Unassigned; grading is the triage seat's call.
The observation
scripts/pm/dispatch-gates.mjscarries a--self-testmode — 28 cases after #8159, covering workflow extraction, script resolution, watch-hint scanning, the runnable-invocation rendering and the new change-kind derivation. No CI job runs it. Measured on currentmain:grep -rn "dispatch-gates" .github/workflows/*.yml— no hits;package.jsonhas nocheck:*script pointing at it (the twoscripts/pm/scripts that DO have gates arecheck:pm-skill-ratchetandcheck:pm-skill-id-lint).So the self-test executes only when a human or agent types it. A change that breaks the extraction functions lands green, and the breakage surfaces as a dispatch prompt quietly naming the wrong gate families — which is the failure this tool exists to prevent, and the kind that reads as correct output.
Why it is not simply the same as #8110
#8110 (
check:skill-frame-freshness --self-testruns nowhere) is the same class but has a different remedy, which is why this is filed beside it rather than as a comment on it:check:skill-frame-freshnessis a CI gate — its invocation exists and the fix is to extend it to--self-test && …, the conventional script shape used everywhere else in this repo;dispatch-gates.mjsis advisory tooling with no CI job at all — there is no invocation to extend, so wiring it needs a deliberate decision about whether a PM-loop tool earns a gate slot, and if so where it hangs.Both are worth deciding together; neither answer implies the other.
Directions (not a recommendation — grading is triage's)
check:pm-dispatch-gatesscript that is--self-testonly (the live derivation reads the workflow tree and would be a slow, noisy gate), wired wherever the two existingscripts/pm/gates run.scripts/pm/**self-test into one job — one step that runs each pm script's--self-test, which also covers whatever lands there next without a per-script decision.Related
check:skill-frame-freshness --self-testruns nowhere in CI, so a broken fixture is invisible until a human runs the gate #8110 — the sibling instance on a script that IS a gate (pm:blockedat time of filing).check-changeset-no-major.mjshas no--self-test, and its enforcing half is unexecuted on every run today (pre-mode) #6923 — prior findings in the same family (self-tests that are exempt from, or absent on, the very job that should run them).