stack/agents/filbert/work/queue-149/r2/body.md at dev-console

0 Fork 1

Files

T

jason.woltje b9ea9304fd lead: commit filbert's untracked evidence (273 files: queue-101 queue-102 queue-105 queue-107 queue-108 queue-109 queue-110 queue-112 queue-113 queue-114 queue-115 queue-116 queue-119 queue-120 queue-122 queue-123 queue-125 queue-126 queue-128 queue-129 queue-130 queue-131 queue-132 queue-133 queue-134 queue-135 queue-136 queue-137 queue-139 queue-140 queue-141 queue-142 queue-145 queue-146 queue-147 queue-148 queue-149 queue-150 queue-151 queue-152 queue-153 queue-154 queue-155 queue-156 queue-162 queue-163 queue-165 queue-166 queue-167 queue-169 queue-170 queue-175 queue-177 queue-181 queue-182 queue-184 queue-185 queue-186 queue-187 queue-188 queue-49 queue-64 queue-66 queue-70 queue-71 queue-80 queue-84 queue-85 queue-87 queue-88)

Unclaimed files left in the canonical checkout. Inventory: canonical-cleanup-inventory.md (commit-as-evidence groups). detect-secrets scan clean on 2026-11-02: hits are sha256 pins, git object ids, test sentinels already in tracked tests and unit-file lines.

2026-11-02 15:05:14 -06:00

5.2 KiB

Raw Permalink Blame History

Filbert review, row 149 round 2: changes.

Candidate: manifest a325a32568a6a2635837455795b0965f1126f31fc25c4048e56e132263aeffcb, Dewey's build.patch applied to ac55e84fe (origin/refactor). In a fresh worktree all 21 hashes check. The patch's added and removed lines are the same as round 1's, so only context moved. I didn't post a round 1 verdict, so this round covers the whole row, not only the 5 drifted paths. Lines below are at the candidate tree.

Changes

merge-gate can't land in the run that raised the decision, and merge-gate.md says it can (roles/merge-gate.md:16-20).
- land_work exists only when the broker's landHeld is true at launch (packages/cli/src/launcher.mjs:262, packages/harness/src/tools.mjs:134). The git socket is bound only then too (packages/cli/README.md:432-434), because merge-gate holds neither git.push.working nor review.verdict. The runner reads tools.json once, at start (packages/harness/src/runner.mjs:322).
- In the raising run, landHeld was false at launch. When cto resolves, the approval comes in as a DECISION message to that same run (packages/bus/src/broker.mjs:1597-1612), and that run has no land_work. The bundle's "Landing:" prompt line (packages/harness/src/bundle.mjs:118) is missing too. So line 16's "Once the decision is resolved ..., call land_work" can't be done.
- Row 137 meant it this way: "row 136 lets a later run of the same instance use it" (brief docs/plans/2026-10-15_q167-rows-r3.md:328-331, blob 606f7e0905). A second run of merge-gate, in the same thread, has to start before anything lands. No contract says who starts it. merge-gate can't message anyone: its RESULT reply to the DECISION message is refused, per your probe. cto.md and chief-of-staff.md say nothing about relaunching it. As shipped, an approved merge sits unused until a person notices.
- Fix:
  - Say the two-run flow in merge-gate.md: land_work appears only in a run launched after the approval, in the thread that raised it. In the raising run, the approval arrives without the tool.
  - Name who starts the landing run in that role's contract, or record it in BUILD.md's Follow-ups with its cost, the way follow-up 1 does for message.send.
  - Add a test that pins the path. In packages/bus/tests/workflow-roles.test.mjs, b.landHeld({ business: 'demo', role: 'merge-gate', thread: null }) should be false before cto resolves, true after, and false once the approval is used.
  - roles/merge-gate.md is test-read (packages/business/tests/workflow-roles.test.mjs:59, 93), so this isn't doc-only under decision 137.

Minor, fix in the same round

The hard-wait list in pm.md:53-56 and chief-of-staff.md:52-55 leaves out one case the broker treats as a hard wait. #hardWait is GATED plus role.revoke aimed at an END_PROTECTED instance, which today is assistant (broker.mjs:139, 516-519). A cited role.revoke of the assistant goes to the human (broker.mjs:1466-1469). The text says "every hard wait goes to the human" and then lists only the eight GATED_ONLY actions. Keep the eight-action sentence, since the test pins it to GATED_ONLY at workflow-roles.test.mjs:112-114. Add a sentence for ending the assistant's run.

Checked and correct

Brief: every "What ships" bullet and every listed test (q167-rows-r3.md:162-216).
- security-reviewer: read, grep, find, ls; review.verdict within role only; no crossRole; no credentials.
- merge-gate: crossRole git.merge.protected only; empty withinRole; no write tools.
- The proactive line is inside Authority in all 13 contracts.
- The domain enum in tools.mjs is technical and delivery, with no default sent.
- pm and chief-of-staff carry the REQ-DEC-2 text, and their old "for the human" wording is gone.
raise_decision forwards domain. The broker defaults it to delivery and refuses anything else with invalid-domain (broker.mjs:1461-1462). The arbitrated route matches the new tool text and the contract text (broker.mjs:1466-1469).
Row 133's refusal works on the shipped lists. merge-gate's receiveFrom includes cto. checkMessaging(route, raiser) runs before anything is stored (broker.mjs:1473). The bus test drops cto and gets decision-route-refused with 0 decisions.
Naming either new role as an arbiter is refused because both declare messaging lists (packages/business/src/business.mjs:167-168). The business test pins that.
security-reviewer's REVIEW-RESULT reaches the coder through review.verdict, with no message.send (bus test lines 86-99). fetch_review is offered only with a checkout (tools.mjs:151). Your follow-up 2 covers a reviewer that has no way to read a ref.
The rebase onto ac55e84fe. Row 143's Duties bullets in assistant, chief-of-staff, pm and cto don't conflict with this row's Authority paragraphs. In tools.mjs only raise_decision changed; MESSAGE_CLASSES keeps BLOCK.
READMEs: the packages/business section and the packages/harness paragraph match the code.

Not run by me

I ran no tests, mutants or gate. Finding 1 changes merge-gate.md and adds a test, so I'll run the touched files singly in round 3 with a slot. Your singles (r2/singles.log, 37 files, 337 tests) and round 1's mutants and gate are your receipts, not mine.

Verdict: changes, on 1. Item 2 is small and can go in the same round.

5.2 KiB Raw Permalink Blame History

Changes

Minor, fix in the same round

Checked and correct

Not run by me

5.2 KiB

Raw Permalink Blame History