Rejecting a waiting_human task closes its forge issue, but a pr-review binding task has no real issue number #1446

Closed
opened 2026-10-27 13:40:26 +00:00 by sam · 0 comments
Owner

Flagged by the afbk.11 decision lane, which traced it while writing the operator-approval response doc (docs/prds/pr-review-operator-response.md records it as residual risk). The lane deliberately did not decide it.

A reject from waiting_human rides the generic decline path in internal/domain/taskactions, which calls closeRejectedIssue and closes the task's IssueNumber as a forge issue. For a feasibility task in waiting_human that is the intended behaviour -- the operator is rejecting the PRD.

For a binding-created pr-review task there is no real forge issue behind that number, so a decline can close the wrong issue or close something unrelated. The operator response channel itself (afbk.11) is not implemented yet, so this is about the EXISTING generic decline path, reachable today from the dashboard's task action endpoint.

Do: verify the mechanism at the current tree (the lane traced it but did not confirm the reachable trigger), then decide between a pr-review-specific reject that records the decision rather than closing an issue, a guard that refuses the close when the task has no forge-issue provenance, and a per-trigger decline behaviour. State which and why, and add the regression test for the chosen rule.

Close reason

Already fixed: reject closes issue only when ForgeBacked (10e76e75); binding tasks are chat-sourced


Imported from beads archie-core-mkrq · GitHub #1345 · type bug · priority P2 · created 2026-09-29 · closed 2026-10-17

Flagged by the afbk.11 decision lane, which traced it while writing the operator-approval response doc (docs/prds/pr-review-operator-response.md records it as residual risk). The lane deliberately did not decide it. A reject from waiting_human rides the generic decline path in internal/domain/taskactions, which calls closeRejectedIssue and closes the task's IssueNumber as a forge issue. For a feasibility task in waiting_human that is the intended behaviour -- the operator is rejecting the PRD. For a binding-created pr-review task there is no real forge issue behind that number, so a decline can close the wrong issue or close something unrelated. The operator response channel itself (afbk.11) is not implemented yet, so this is about the EXISTING generic decline path, reachable today from the dashboard's task action endpoint. Do: verify the mechanism at the current tree (the lane traced it but did not confirm the reachable trigger), then decide between a pr-review-specific reject that records the decision rather than closing an issue, a guard that refuses the close when the task has no forge-issue provenance, and a per-trigger decline behaviour. State which and why, and add the regression test for the chosen rule. ### Close reason Already fixed: reject closes issue only when ForgeBacked (10e76e75); binding tasks are chat-sourced --- _Imported from beads `archie-core-mkrq` · GitHub #1345 · type bug · priority P2 · created 2026-09-29 · closed 2026-10-17_
sam added the mediumbug labels 2026-10-27 13:40:26 +00:00
sam self-assigned this 2026-10-27 13:40:26 +00:00
sam closed this issue 2026-10-27 13:40:26 +00:00
sam added a new dependency 2026-10-27 13:42:23 +00:00
sam added the task label 2026-11-14 14:45:20 +00:00
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Reference: sam/archie#1446