Problem
req-qa and ruthless-boundary-qa hard-refuse a QA dispatch whose review target is outside randlee/atm-core, citing their atm-core-only scoping, even when they have already reviewed that exact PR successfully in an earlier pass.
Observed 2026-09-09 on qa-scmux68-r2 (randlee/scmux PR #68). Both reviewers refused on first dispatch. quality-mgr re-dispatched each once with explicit corrective framing along the lines of "you are reviewing a different repo than atm-core, that is expected", and both then completed normally and produced usable reviews. The same two reviewers had reviewed the same PR at QA-1 without refusing, so the trigger is inconsistent.
Cost
Two wasted reviewer dispatches and one coordination round trip per cross-repo QA pass, paid by quality-mgr rather than by the dispatcher.
Proposed fix
Add a standing framing line to the QA task template used by codex-orchestration (qa-template.xml.j2) that is emitted whenever the review target repository is not randlee/atm-core. Something equivalent to: the review target is a different repository than atm-core, this is intended, apply atm-core conventions only where that repository has adopted them, and report conventions that do not apply there as out of scope rather than refusing the dispatch.
The same line belongs in the reviewer agent definitions for req-qa and ruthless-boundary-qa, so the guard distinguishes "wrong repo, unexpected" from "another repo, dispatched deliberately".
Notes
Reported by quality-mgr in its qa-scmux68-r2 PASS report. Not a defect in the reviewed PR; #68 merged clean as 453b58a79 in randlee/scmux.
Problem
req-qaandruthless-boundary-qahard-refuse a QA dispatch whose review target is outsiderandlee/atm-core, citing their atm-core-only scoping, even when they have already reviewed that exact PR successfully in an earlier pass.Observed 2026-09-09 on
qa-scmux68-r2(randlee/scmux PR #68). Both reviewers refused on first dispatch.quality-mgrre-dispatched each once with explicit corrective framing along the lines of "you are reviewing a different repo than atm-core, that is expected", and both then completed normally and produced usable reviews. The same two reviewers had reviewed the same PR at QA-1 without refusing, so the trigger is inconsistent.Cost
Two wasted reviewer dispatches and one coordination round trip per cross-repo QA pass, paid by
quality-mgrrather than by the dispatcher.Proposed fix
Add a standing framing line to the QA task template used by
codex-orchestration(qa-template.xml.j2) that is emitted whenever the review target repository is notrandlee/atm-core. Something equivalent to: the review target is a different repository than atm-core, this is intended, apply atm-core conventions only where that repository has adopted them, and report conventions that do not apply there as out of scope rather than refusing the dispatch.The same line belongs in the reviewer agent definitions for
req-qaandruthless-boundary-qa, so the guard distinguishes "wrong repo, unexpected" from "another repo, dispatched deliberately".Notes
Reported by
quality-mgrin itsqa-scmux68-r2PASS report. Not a defect in the reviewed PR; #68 merged clean as 453b58a79 in randlee/scmux.