Distillation

🟡 (02:15) Assistant identified the parser conflict and chose one canonical implementation: approval-authorizer.js keeps a fail-closed parsePublishTitle() wrapper, publication keeps precise validation errors, both use the generated grammar plus shared validators, and PUBLISH_TITLE_REGEX is removed entirely.
🔴 (02:15) Files updated while resolving shared-module conflicts: src/modules/details-from-context.js, src/publish/inputs.js, src/modules/update-issue.js, src/modules/__tests__/details-from-context.js, and src/modules/__tests__/update-issue.js.
🟡 (02:15) Assistant stated the resolved shared-module behavior now uses the generated title parser, an explicit workflow-provided state-file path, and unconditional approval cleanup.
🔴 (02:16) After resolving shared modules, merge-conflict markers remained only in .github/workflows/publish.yml at lines 192–196 and 261–265, and .github/workflows/ci-poller.yml at lines 131–193.
🔴 (02:16) Focused Vitest command used vitest run against 11 files: src/modules/__tests__/details-from-context.js, src/modules/__tests__/approval-authorizer.js, src/modules/__tests__/update-issue.js, src/modules/__tests__/publish-issue-title.js, src/modules/__tests__/release-revision.js, src/modules/__tests__/publish-location.js, src/modules/__tests__/ci-poller-input.js, src/publish/__tests__/authorize-approval.js, src/publish/__tests__/resolve-release-revision.js, src/publish/__tests__/resolve-location.js, and src/publish/__tests__/discover-location.js, with --printConsoleTrace.
🔴 (02:16) Focused tests passed 96 of 97 tests. The sole failure was uses the shared resolver in the CI poller in src/modules/__tests__/ci-poller-input.js; that file had 14 tests total with 13 passing and 1 failing because .github/workflows/ci-poller.yml was still unresolved.
🔴 (02:16) Individual displayed passing results included src/modules/__tests__/approval-authorizer.js (15 tests), src/modules/__tests__/update-issue.js (7 tests), src/publish/__tests__/authorize-approval.js (7 tests), src/modules/__tests__/details-from-context.js (17 tests), src/modules/__tests__/release-revision.js (13 tests), and src/publish/__tests__/resolve-release-revision.js (1 test).
🔴 (02:16) User specified for workflow_dispatch: “Always run trusted code,” because workflow_dispatch can target any ref.
🔴 (02:16) User specified that Publish’s title parser must be used so CI and publishing “always target” the same canonical publish request rather than independently interpreting the title.
🔴 (02:16) User specified: “Never move a release to ci-ready after it changes.”
🔴 (02:16) User specified that a renamed or re-approved issue “never reaches ci-ready” unless its approval is revalidated against the current canonical request.
🔴 (02:16) The proposed CI-poller workflow uses the sentry-internal-app token for label changes, warns and removes accepted when an issue’s approval is invalid, parses titles through Publish’s parser, and derives the repository as "getsentry/" + .repo.
🔴 (02:16) The proposed CI-poller workflow treats zero check suites as a reason to use the issue SHA, counts checks with non-successful conclusions, and includes self-dispatch for fast re-checking while issues remain pending.
🟡 (02:16) Assistant chose to resolve .github/workflows/ci-poller.yml by combining canonical input parsing with approval binding. If the canonical SHA changes during inspection, the poller will stop that cycle and revoke approval instead of silently publishing the mutated request.

Distillation

Observations