[qingyun-wu]

main is red, and this is why

tests/worker-delivery-matches-the-hook.test.py asserts the Stop hook still defines sentinel_task_id(). #4437 deliberately removed it when the hook started shelling out to src/delivery/task_dispatch.py. Measured on origin/main (f7a553eab) just now, unmodified checkout:

$ python3 tests/worker-delivery-matches-the-hook.test.py
  FAIL the hook still defines sentinel_task_id()
FAILED: the hook still defines sentinel_task_id()

Two consecutive main CI runs are red on it (f7a553eab, 82d3b7093), so every PR whose python standalone tests job runs inherits the failure — that is the red on #4466 as well, which is otherwise clean (@john-the-dev reached the same conclusion from the import graph; this is the measurement).

The duplication did not disappear, it moved

The suite exists because the suffix set is spelled in more than one place and "the copy nobody re-reads is the one that hands a worker's task back to the core". After #4437 the hook holds no copy — but a new one appeared:

owner	spelling
skills/worker-pool/scripts/worker_delivery.py	SENTINEL_SUFFIXES (from pool_delivery)
src/delivery/task_dispatch.py	_WORKER_HOLD_SUFFIXES = (".txt", ".accepted", ".claimed")

They agree today (['.accepted', '.claimed', '.txt'] both). They cannot share the constant: src/ must not import an optional skill (CLAUDE.md, "Optional capability discovery stays at the adapter edge"), so this suite is the only thing keeping them in step — and it was pinned to the hook, which no longer participates.

What the guard asserts now

• the hook spells no sentinel suffix of its own (it delegates);
• the hook asks task_dispatch.py owned-by for ownership;
• the two module spellings are the same set.

Mutation evidence

Each assertion mutated separately, reverted by reverse replace:

mutation	result
add case "$1" in *.accepted) :;; esac to the hook	FAIL the hook spells no sentinel suffix of its own (found ['.accepted'])
break the task_dispatch.py owned-by call	FAIL the hook asks task_dispatch.py owned-by for ownership
drop ".claimed" from _WORKER_HOLD_SUFFIXES	FAIL task_dispatch ['.accepted', '.txt'] == worker_delivery ['.accepted', '.claimed', '.txt']

A first, looser form of the delegation check ("task_dispatch.py" in hook) survived its mutation, because the hook names the module three times — it is anchored on the owned-by call for that reason, and the table above is the re-run.

After: all checks passed, 23 checks, same file. review-checks.sh --diff PASS; lint-workspace-resolution.sh --diff clean.

Test-only; no production behaviour changes. Single concern, no bundled refactor.

🤖 Generated with Claude Code

[qingyun-wu]

Closing in favour of #4467, which was opened 80 seconds earlier (18:47:47Z vs 18:49:07Z) and fixes the same red — first PR opened wins the claim.

Verified #4467 green at 5576095d and approved it there. The one thing this branch had that #4467 does not is the module-vs-module pin: dropping ".claimed" from task_dispatch._WORKER_HOLD_SUFFIXES leaves #4467's suite fully green, so the two live spellings (worker_delivery.SENTINEL_SUFFIXES and _WORKER_HOLD_SUFFIXES) are unguarded. I have named it on #4467 as a non-blocking follow-up and will open it separately once that lands, so the RC is not held up.

Branch kept for that follow-up.

[github-actions]

Coverage Gate

✅ Diff coverage PASSES the 95% bar. Whole-tree (informational): 89%.

Diff Coverage

Diff: origin/main...HEAD, staged and unstaged changes

No lines with coverage information in this diff.