Retrospective finding — refs #10.
Context
Two related findings:
- Traceability is near zero. Across roughly 260 commits in eight single-maintainer repos,
effectively none reference an issue — the only refs present were bot/external-PR artifacts.
One repo demonstrably ran issue-driven work (its changelog cites issue numbers) yet not one
commit subject carries them: the traceability lives in changelog prose where no tooling can
follow it. The standard mandates Conventional Commits but says nothing about issue linkage.
- Auto-close keywords are a footgun where a human gate exists. In a multi-person,
board-driven project, the documented Closes #NN convention auto-closed seven issues at PR
merge — silently skipping the project's human verification step, which by design happens
after landing on the main branch. The local fix: reference issues as plain Refs #NN,
humans close after verifying, and a CI job rejects any PR whose body or commit messages
contain a closing keyword followed by an issue reference.
Proposal
- Convention: when a repo uses an issue tracker, commits/PRs reference the driving issue as
Refs #NN (subject or footer). Repos without a tracker are exempt — but must not half-adopt
(issue numbers in changelog prose only is the worst of both).
- Closing-keyword guidance: closing keywords are appropriate only where merge genuinely
equals done (solo repo, no post-merge verification). Where any human verify step follows the
merge, the standard prescribes Refs + manual close.
- Ship the guard: provide the closing-keyword-rejecting CI job as a reusable snippet in the
standard (it exists and is field-tested), plus a PR-template line prompting Refs #NN.
Tasks
Acceptance criteria
- A repo adopting the standard can answer "which issue drove this commit" from
git log alone.
- The auto-close failure mode is documented with its guard available as copy-paste.
— 🤖 Claude, on behalf of @PBNZ
Retrospective finding — refs #10.
Context
Two related findings:
effectively none reference an issue — the only refs present were bot/external-PR artifacts.
One repo demonstrably ran issue-driven work (its changelog cites issue numbers) yet not one
commit subject carries them: the traceability lives in changelog prose where no tooling can
follow it. The standard mandates Conventional Commits but says nothing about issue linkage.
board-driven project, the documented
Closes #NNconvention auto-closed seven issues at PRmerge — silently skipping the project's human verification step, which by design happens
after landing on the main branch. The local fix: reference issues as plain
Refs #NN,humans close after verifying, and a CI job rejects any PR whose body or commit messages
contain a closing keyword followed by an issue reference.
Proposal
Refs #NN(subject or footer). Repos without a tracker are exempt — but must not half-adopt(issue numbers in changelog prose only is the worst of both).
equals done (solo repo, no post-merge verification). Where any human verify step follows the
merge, the standard prescribes
Refs+ manual close.standard (it exists and is field-tested), plus a PR-template line prompting
Refs #NN.Tasks
Acceptance criteria
git logalone.— 🤖 Claude, on behalf of @PBNZ