docs(consult): make the label track the thread, since nothing else does #439

Merged

coilyco-ops merged 1 commit from docs/the-consult-label-tracks-the-thread into main

2026-08-13 12:58:47 +00:00

coilyco-ops commented

2026-08-13 12:56:15 +00:00

Member

closes #438

The structural half of #437. Darren corrected the drift by hand and said what happens next:

The queue is currently honest. It will drift again by tomorrow without one of the fixes above, because nothing structural changed.

The measurement, because the rule is uninteresting without it

The queue advertised 18 items when 8 were real, while hiding 5. A reader who trusted it would have been wrong about 15 of 23.

One of those ten was mine. Answered at 07:57Z, not picked up for four hours, while I worked other issues — because the board said blocked and I believed it. That is the cost, and it is not inattention.

Two habits, each one call

recording a decision removes consult, in the same call that records it
asking a human a question adds it, in the same call that asks

Both attach to a write that is already happening. They go in AGENTS.md, beside the rules that already bind PR bodies and prose style, because that is what makes them binding on the next agent rather than on whoever happened to read the thread.

Why not a check

Deciding whether a comment holds an unanswered question is a judgement, not a pattern. Both directions are easy to detect approximately and hard to detect correctly — and a hygiene check that is wrong sometimes trains people to ignore it, which is the same failure mode as a permanently red security row.

Darren's flag-rather-than-enforce sweep is the reasonable middle. It is recorded in the doc as worth building, and it is not a prerequisite for the habits.

ward exec gate green: build, policy-check, vet, test, test-skips, pre-commit.

closes #438 The structural half of https://forgejo.coilysiren.me/coilyco-gaming/sirens-echo/issues/437. Darren corrected the drift by hand and said what happens next: > The queue is currently honest. **It will drift again by tomorrow** without one of the fixes above, because nothing structural changed. ## The measurement, because the rule is uninteresting without it The queue advertised **18** items when **8** were real, while hiding **5**. A reader who trusted it would have been wrong about 15 of 23. **One of those ten was mine.** Answered at 07:57Z, not picked up for four hours, while I worked other issues — because the board said blocked and I believed it. That is the cost, and it is not inattention. ## Two habits, each one call - recording a decision **removes** `consult`, in the same call that records it - asking a human a question **adds** it, in the same call that asks Both attach to a write that is already happening. They go in `AGENTS.md`, beside the rules that already bind PR bodies and prose style, because that is what makes them binding on the next agent rather than on whoever happened to read the thread. ## Why not a check Deciding whether a comment holds an *unanswered* question is a judgement, not a pattern. Both directions are easy to detect approximately and hard to detect correctly — and a hygiene check that is wrong sometimes trains people to ignore it, which is the same failure mode as a permanently red security row. Darren's flag-rather-than-enforce sweep is the reasonable middle. It is recorded in the doc as worth building, and it is not a prerequisite for the habits. `ward exec gate` green: build, policy-check, vet, test, test-skips, pre-commit.

coilyco-ops added 1 commit

2026-08-13 12:56:16 +00:00

docs(consult): make the label track the thread, since nothing else does

ci / image-build (pull_request) Successful in 19s

Details

ci / test (pull_request) Successful in 40s

Details

ci / publish-echo-image (pull_request) Has been skipped

Details

ci / publish-observed (pull_request) Has been skipped

Details

7855752601

consult decides how cli-guard dispatches an issue and what a human sees when
they look for work that needs them. It is written once and never reconciled,
so it drifts both ways: an answer arrives and the label stays, or a question is
asked in prose and no label is written.

The second is worse. Unlabelled is correctly fail closed for dispatch and
invisible to the human, so the issue is both undispatched and unqueued.

Measured on one sweep: the queue advertised eighteen items when eight were
real, while hiding five. A reader who trusted it would have been wrong about
fifteen of twenty-three. One of the ten was mine, answered at 07:57 and not
picked up for four hours, because the board said blocked and I believed it.

Both habits attach to a write that is already happening, so they cost one call
and nothing else. They go in AGENTS.md rather than a thread, which is what
makes them binding on the next agent rather than on whoever read the thread.

Not a check. Deciding whether a comment holds an unanswered question is a
judgement rather than a pattern, and a hygiene check that is wrong sometimes
trains people to ignore it. The flag-rather-than-enforce sweep is recorded as
worth building and is not a prerequisite.

closes #438

Co-authored-by: Kai Siren <coilysiren@gmail.com>
Co-authored-by: Claude <noreply@anthropic.com>

coilyco-ops requested review from coilysiren

2026-08-13 12:56:16 +00:00

coilyco-ops referenced this pull request

2026-08-13 12:56:20 +00:00

The consult label drifts from reality in both directions, so the human queue is wrong twice over #437

coilyco-ops merged commit 58443f74c7 into main

2026-08-13 12:58:47 +00:00

coilyco-ops referenced this pull request from a commit

2026-08-13 12:58:49 +00:00

Merge pull request 'docs(consult): make the label track the thread, since nothing else does' (#439) from docs/the-consult-label-tracks-the-thread into main

coilyco-ops referenced this pull request

2026-08-13 18:58:42 +00:00