Summary
The operate skill says lane brief requires a closing issue link, but the CLI supports Part of PRs. A paper-only operator evaluation repeated the outdated restriction.
What I was doing
Testing whether the existing Fabrika operator could recover after a workflow host disappeared while its builder completed, without executing any live lane actions.
What I observed
In claude-plugins/fabrika/skills/operate/SKILL.md section 2, the lane brief paragraph says it counts a PR only when it declares it closes the issue. However packages/fabrika-cli/src/lane/brief-verb.unit.test.ts:502 explicitly tests resolving a Part of PR that the closing edge cannot see. packages/fabrika-cli/src/lane/prove-verb.unit.test.ts:701 tests the corresponding DONE proof.
Operator evaluation child a481348a-a4d5-4581-97c4-413a0cb897b4, workflow 15961e0d-54da-4612-98f1-a0890696445c, answered the no-ledger partial-PR case: "Part of alone does not satisfy that contract." That follows the prose but disagrees with the tested CLI. Existing brief/report/prove tests passed: 140 tests across three files. This was a paper-only model evaluation, not live command execution.
Why it matters
An operator can reject a supported PR relationship and report a false blocker. This is separate from the real scope limit: an issue lane returns to queued after a partial merge, so authorizing completion of one PR does not authorize implementing the rest of the issue.
Pointers
Suggested next step (non-binding)
Align the authored operate guidance and published Pi bundle with the supported Part of nomination, and retain a regression evaluation that distinguishes PR linkage from PR-only authorization. Do not remove the partial-merge continuation guard. Confidence 99% in the documentation mismatch.
Filed by an agent · session 425cb3d4-b7c2-4bc3-b562-0776c88e7787 · branch umut/tuval-pi-0.85.0 · 2026-09-08T20:55:30Z
Summary
The operate skill says lane brief requires a closing issue link, but the CLI supports Part of PRs. A paper-only operator evaluation repeated the outdated restriction.
What I was doing
Testing whether the existing Fabrika operator could recover after a workflow host disappeared while its builder completed, without executing any live lane actions.
What I observed
In claude-plugins/fabrika/skills/operate/SKILL.md section 2, the lane brief paragraph says it counts a PR only when it declares it closes the issue. However packages/fabrika-cli/src/lane/brief-verb.unit.test.ts:502 explicitly tests resolving a Part of PR that the closing edge cannot see. packages/fabrika-cli/src/lane/prove-verb.unit.test.ts:701 tests the corresponding DONE proof.
Operator evaluation child a481348a-a4d5-4581-97c4-413a0cb897b4, workflow 15961e0d-54da-4612-98f1-a0890696445c, answered the no-ledger partial-PR case: "Part of alone does not satisfy that contract." That follows the prose but disagrees with the tested CLI. Existing brief/report/prove tests passed: 140 tests across three files. This was a paper-only model evaluation, not live command execution.
Why it matters
An operator can reject a supported PR relationship and report a false blocker. This is separate from the real scope limit: an issue lane returns to queued after a partial merge, so authorizing completion of one PR does not authorize implementing the rest of the issue.
Pointers
Suggested next step (non-binding)
Align the authored operate guidance and published Pi bundle with the supported Part of nomination, and retain a regression evaluation that distinguishes PR linkage from PR-only authorization. Do not remove the partial-merge continuation guard. Confidence 99% in the documentation mismatch.
Filed by an agent · session
425cb3d4-b7c2-4bc3-b562-0776c88e7787· branchumut/tuval-pi-0.85.0· 2026-09-08T20:55:30Z