What happened
Four Dependabot pull requests went red on conformance / check after gh pr update-branch, on a
commit nobody wrote by hand:
FAIL 006cc53 no Assisted-by: trailer as its final lines
skip 78502ea dependabot[bot] is a bot — its commits are generated and cannot carry a trailer
006cc53 is the merge commit update-branch writes — two parents, message
Merge branch 'main' into dependabot/..., authored by whoever pressed the button.
Why it is worth recording rather than working around
CONTRIBUTING.md says, of the commits GitHub writes:
the Merge branch 'main' into … commit that the Update branch button writes when your branch
has fallen behind. The check skips a merge commit and is never run over main, so neither is
yours to get right
On this evidence the check does not skip a merge commit. It skips a commit whose author is a
bot — the skip line above says so in its own words — and an Update branch merge is authored by
the person who pressed it, so it is judged as theirs and fails for a trailer it structurally
cannot carry.
The cost is not theoretical: Dependabot then refuses to rebase the branch, because it has been
"edited by someone other than Dependabot", and the way back is @dependabot recreate on every
affected pull request.
Whose to fix
The check is ../dot-github/conformance/check-trailers.sh — the Organisation's, arriving written,
and not this repository's to own (CODING_STANDARDS.md §5). So this issue is a report, and the
decision is which of the two is wrong:
- the check, if the intent is what
CONTRIBUTING.md describes, in which case it should skip
any commit with two parents; or
- the wording, if skipping by author is the intent, in which case
CONTRIBUTING.md and
AGENTS.md describe behaviour the check has never had.
Either way both encodings should say the same thing, which is the argument those files already make
about the two encodings of the auto-merge conditions.
Acceptance criteria
Assisted-by: Claude Opus 5
Co-authored-by: Claude Opus 5 noreply@anthropic.com
What happened
Four Dependabot pull requests went red on
conformance / checkaftergh pr update-branch, on acommit nobody wrote by hand:
006cc53is the merge commitupdate-branchwrites — two parents, messageMerge branch 'main' into dependabot/..., authored by whoever pressed the button.Why it is worth recording rather than working around
CONTRIBUTING.mdsays, of the commits GitHub writes:On this evidence the check does not skip a merge commit. It skips a commit whose author is a
bot — the
skipline above says so in its own words — and an Update branch merge is authored bythe person who pressed it, so it is judged as theirs and fails for a trailer it structurally
cannot carry.
The cost is not theoretical: Dependabot then refuses to rebase the branch, because it has been
"edited by someone other than Dependabot", and the way back is
@dependabot recreateon everyaffected pull request.
Whose to fix
The check is
../dot-github/conformance/check-trailers.sh— the Organisation's, arriving written,and not this repository's to own (
CODING_STANDARDS.md§5). So this issue is a report, and thedecision is which of the two is wrong:
CONTRIBUTING.mddescribes, in which case it should skipany commit with two parents; or
CONTRIBUTING.mdandAGENTS.mddescribe behaviour the check has never had.Either way both encodings should say the same thing, which is the argument those files already make
about the two encodings of the auto-merge conditions.
Acceptance criteria
CONTRIBUTING.mdandAGENTS.mdhere describe what the check actually does.update-branchindocs/agents/dependencies.mdis removed or kept on purpose, depending on which way it went.Assisted-by: Claude Opus 5
Co-authored-by: Claude Opus 5 noreply@anthropic.com