fix(record): find the trailer block where the merge tool left it (#406) - #409
Merged
Conversation
main is red on agent-record, and one of the five failing shapes is the gate's own fault: it rejects a commit that is entirely correct. check-commit-trailers.py reads trailers through `git interpret-trailers --parse`, which by design treats ONLY the final paragraph as the trailer block. GitHub appends `Co-authored-by:` as a SEPARATE trailing paragraph on a squash merge, so a complete correct block becomes invisible. Piping `git show -s --format=%B dbd0d51` into `git interpret-trailers --parse` prints nothing but that one co-author line, and the gate duly reports trailers the commit plainly carries as missing. 13 of the last 30 commits on main fail this check -- unnoticed only because those runs were cancelled (#274), which HID the defect rather than causing it. The fix fuses consecutive trailing TRAILER-SHAPED paragraphs before parsing. Nothing is relaxed: the block must still exist, the marker must still sit above it, each declaration must still appear exactly once, and an AI co-author is still forbidden. A prose paragraph still terminates the block, so trailers buried mid-message stay invalid. I got this wrong first and the reverted half is the more instructive one. The first attempt ALSO collapsed identical duplicate trailers, to fix the multi-commit-squash shape where the whole block appears twice. That relaxes the uniqueness rule, and an existing test already pins it -- rewriting that assertion to suit my change is exactly what AGENTS.md forbids. The distinction is real: a doubled block is a genuinely malformed message, fixable at source by writing the squash body or landing a single-commit PR, whereas the Co-authored-by case is a correct commit defeated by the parser. Reverted in full; b8293c8 stays red on purpose. SCOPE LIMIT, stated rather than implied: this fixes ONE of five observed shapes, and does NOT by itself make main green. Verified per commit -- dbd0d51 human co-author appended PASS (was failing; the fix) f64f2b7 BOT co-author appended fails: a REAL violation the parse had been hiding, now correctly surfaced 87308de GitHub's `---------` separator fails: malformed b8293c8 squash doubled the block fails: malformed b580452 merge button, no trailers fails: real violation The remaining four are merge-method artifacts. Closing them is a process change, not a checker change, and the spec names it as follow-up. Gates: preflight rc=0; test_check_commit_trailers 21 (the RED-BEFORE case plus four guards that keep the fusion bounded -- doubled block, contradictory declarations, no-trailer merge message, and prose after the block -- all green before and after); test_check_gate_commands 32. Issue: #406 FOLLOWING_AGENTS_PROTOCOL Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:claude-opus-5 [Claude Code]
This was referenced Aug 11, 2026
localai-bot
pushed a commit
that referenced
this pull request
Aug 11, 2026
…418) check-pr-size correctly refused the change: scripts/check-agent-record.py moved without added evidence in its PAIRED suite. The bump is only a data re-pin, but the gate does not distinguish that, and it is right not to -- a bump with no row behind it looks exactly like a bump for a new row, which is the failure the pin exists to catch. Adds two cases keyed to THIS row rather than to the constant in general: the row appears exactly once in the engine matrix, and moving the pin by one must make the count disagree. The second is the mutation; without it the first would pass against a decorative constant. Not fixed here, and stated rather than left implicit: documentation-checkpoint also reds this PR, because adding a row as ACTIVE is a lifecycle claim that owes docs/STATUS.md and docs/BENCHMARKS.md. For a records-and-checker row those pages are the wrong home -- STATUS is the per-capability surface and a commit-trailer policy is not a capability -- so writing a line there to satisfy the gate would be dishonest documentation. #409 and #416 both landed with that same gate red. That is a real gap in the gate rather than in this change, and it deserves its own issue instead of a paper-over here. Gates: test_agent_record 31; check-pr-size now OK over this PR's range. Issue: #418 FOLLOWING_AGENTS_PROTOCOL Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:claude-opus-5 [Claude Code]
localai-bot
pushed a commit
that referenced
this pull request
Aug 11, 2026
… reproduce (#349) Records the canonical six-point regrid: the published 27B c1 figure does not reproduce, and the record now says so rather than carrying a number that only held once. The branch was red for three reasons, none of them its own measurement: check-public-doc-tables conflicted because the branch re-pinned the STATUS `chars` ratchet -- a key #364 deleted, because a byte count of one file stored in another moved on every edit and coupled every PR to lines it did not own. Taking main's version loses nothing: the re-pin existed only to pay for its own STATUS edit. .agents/NOW.md conflicted because the branch edited the per-row live-claims table, which #374 deleted when the live position became derived. The measurement itself was never in NOW.md -- it lives in docs/BENCHMARKS.md and .agents/benchmark-record.md, both of which merged cleanly. .agents/benchmark-record.md is genuinely append-only, so both sides were UNIONED rather than one chosen. That is the one file where union is the correct resolution under AGENTS.md, and taking either side alone would have silently dropped a measurement. Merged --no-ff rather than squashed so the individual record commits stay inspectable, and because a squash of a multi-commit PR concatenates every commit's trailer block and reds the gate -- the one merge-method shape #409 deliberately did not paper over. Gate: scripts/agent-preflight.sh rc=0 on the merged tree. CI is queue-blocked (#274), so the operator's own gate run is the authority, as AGENTS.md prescribes. No product source is touched. FOLLOWING_AGENTS_PROTOCOL Following-Agents-Protocol: true AI-Assisted: true Assisted-by: AGENT:claude-opus-5 [Claude Code]
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Issue #406. Row
ENG-TRAILER-MERGE-ARTIFACTS. Spec.agents/specs/trailer-merge-artifacts.md. Related: #274.This is why
mainis red onagent-record— and one of the five failing shapes is the gate's own fault: it rejects a commit that is entirely correct.Root cause
check-commit-trailers.pyreads trailers viagit interpret-trailers --parse, which treats only the final paragraph as the trailer block. GitHub appendsCo-authored-by:as a separate trailing paragraph on squash merge, so a complete, correct block becomes invisible:13 of the last 30 commits on
mainfail this check — unnoticed only because those runs were cancelled (#274). The cancellation hid the defect; it didn't cause it.The fix
Fuse consecutive trailing trailer-shaped paragraphs before parsing. Nothing is relaxed: the block must still exist, the marker must still sit above it, each declaration must still appear exactly once, an AI co-author is still forbidden, and a prose paragraph still terminates the block.
I got this wrong first, and the reverted half matters more
My first attempt also collapsed identical duplicate trailers, to fix the multi-commit-squash shape. That relaxes the uniqueness rule — and an existing test already pins it. Rewriting that assertion to suit my change is exactly what
AGENTS.mdforbids.The distinction is real: a doubled block is a genuinely malformed message, fixable at source by writing the squash body or landing a single-commit PR; the
Co-authored-bycase is a correct commit defeated by the parser. Reverted in full.Scope limit — this does NOT make main green on its own
dbd0d51cf64f2b7187308dea---------separatorb8293c88b580452dThe other four are merge-method artifacts. Closing them is a process change, not a checker change, and it's the open question I'd like your call on: either always land with an author-written message (which forbids the merge button), or accept that bot co-authors and squash separators will keep redding main.
Evidence
preflight rc=0;
test_check_commit_trailers21 — the RED-BEFORE case plus four guards that keep the fusion bounded (doubled block, contradictory declarations, no-trailer merge message, prose after the block), all four green before and after;test_check_gate_commands32. No product source touched.🤖 Generated with Claude Code