You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
fix(ci): review lanes give no verdict on a diff line over 1800 bytes, holding 371 tracked lines' edits for a person #12263
pr-review-main stops with reason=long-line when any diff line is over LONG_LINE_BYTES (1800; .github/workflows/pr-review-main.yml:436, .github/scripts/pr-review-main/write-verdict.mjs:31). The stop leaves no verdict, so claude-review-status fails, and ci-status's "Require both review lanes on this head" step fails the PR. The merge gate then holds it for a person as long-line-review:<path> (docs/conventions/pr-pipeline/README.md:1259-1263).
A diff carries the removed line as well as the added one. So any PR that edits a line that is already over 1800 bytes on main trips the stop, however the edit is worded. On origin/main at 37e7868, 371 lines in 136 tracked .md/.json/.yaml/.txt files are over 1800 bytes. Examples: docs/contracts/detector-findings/README.md (14 lines), docs/upstream/cursor-pstack.md (23), the table rows in docs/contracts/config-cascade/README.md, and many eval samples/*.json answers.
Seen on #12253 (#11969). The issue's task list requires editing the review row of docs/contracts/config-cascade/README.md. That row is 1954 bytes on main and 2056 after the edit. claude-review stopped long-line on both heads (13f7c90, c10a511), read 18 of 30 files with 0 findings, and ci-status failed. claude-security-review read 30 of 30 with 0 findings.
Why it matters
The owner's intent is that an item stops for a person only on an intent mismatch or a security issue in its PR. This hold stops scoped, agent-ready work for neither reason, so it is a workflow-gate defect.
Fix options
The diff step folds an over-long line into pieces under the cap before splitting into parts (a continuation marker the prompt explains), so Read returns each piece whole and the lane can still give a verdict. This removes the cause for every file. Recommended. Basis: the stop exists only because Read may cut a long line short (pr-review-main.yml:432-435), and folding keeps every byte readable.
Reflow the long lines in tracked docs. This does not cover generated or sample JSON, and each reflow PR trips the stop once.
Keep the hold. Each PR touching one of the 371 lines then waits for the owner.
Done when
A PR whose diff holds a line over 1800 bytes gets a claude-review verdict that ci-status accepts, with every byte of that line read.
A test in .github/scripts/pr-review-main/workflow.test.mjs covers a diff with a 4000-byte line.
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Medium
Complexity: Medium
Uncertainty: Known
Flags: none
Blocked by: none
Understood: pr-review-main stops with reason=long-line on any diff line over 1800 bytes, so no review verdict is posted and ci-status fails. Any PR that edits one of the 371 existing over-long lines is held for a person. #12253 is an example. The issue recommends folding long lines into pieces under the cap, with a continuation marker explained in the prompt, so every byte stays readable. Done-when: a verdict that ci-status accepts for a diff with a long line, and a workflow test with a 4000-byte line. The issue names its own recommended approach, so I found no open intent question. I found no open issue that duplicates this one.
The work-resolve-issue lane stopped on this issue (patch-protected-path): the change needs a path the issue lane may not push, which a maintainer session builds by hand.
It added host-run and removed agent-ready: an agent run on a host builds it, and it waits on no decision from a person.
In the run's words:
The run stopped early (patch-protected-path): its change needs a path the issue lane may not push, which a person changes instead. In the run's words: patch-protected-path: .github/workflows/pr-review-main.yml. The long-line stop and the part split the fix must change sit inline in the 'Write the diff' step of each review leg (the awk stop at lines 455-457, 1279-1281 and 2107-2109, and the part-split awk after each). This lane cannot push changes to /.github/workflows/**. Folding lines in write-diff.py instead would leave the workflow's stop as dead code and its comments wrong, which works around the protected path.
What the run says it did not do:
Both done-when items (a claude-review verdict ci-status accepts on a diff line over 1800 bytes; a workflow.test.mjs test with a 4000-byte line) need an edit to the protected .github/workflows/pr-review-main.yml
Did not check whether the merge gate's long-line-review hold (babysit_risk LONG_LINE_BYTES in plugins/source-control) needs a matching change once the review folds long lines
Repository checks and tidy not run: no file changed
Correction to the Observed section: on #12253 the long-line stop did not fail ci-status.
ci-status failed on pr-review / claude-review-status (run 41386160944). That is the pull_request claude-review lane, and it did run: it posted a partial review that read 18 of 30 changed files. The 12 unread files were whole-file deletions, which its coverage count includes. The hold on refactor(review): remove /review:audit-enforceability #12253 (reason lane-defect) names that cause, with a --diff-filter=d fix in ci-workflows' pr-review.
The long-line stop is a separate record, the pr-review-main: claude-review check run, concluded neutral with verdict=none reason=long-line. That is the main-run review. Per docs/conventions/pr-pipeline/README.md:1259-1263, it holds a PR for a person as long-line-review:<path> only when the PR is one "only the main-run review binds". This issue's claim that it fails ci-status for every PR is wrong.
What stands: the main-run review gives no verdict on a diff holding a line over 1800 bytes, and 371 lines in 136 tracked files on main are over it. That matters for every PR the main-run review alone binds. The folding fix in option 1 still applies to those PRs. The scope is narrower than first written.
Host run. Built the fold that lets the main-run review read diff lines over 1800 bytes, and removed the long-line-review merge hold.
What ran (Claude Code 2.1.316):
The code is in this repository: the long-line stop was inline in the "Write the diff" step of each of the three legs of .github/workflows/pr-review-main.yml, and the person hold was _long_line_reasons in plugins/source-control/skills/babysit-prs/scripts/babysit_risk.py.
Each leg's part split now folds a line over LONG_LINE_BYTES onto \ continued: lines, never inside a UTF-8 character. This is the same fold as pr-verify's chunk-diff.sh.
The prompt says a continuation line adds no line to the file's numbering, so finding line numbers stay correct.
write-verdict.mjs scans model text with its folds joined for credentials.
The hold and its person prefix are removed. The binder's max_line_bytes cap is unchanged.
Tests run:
node --test on workflow.test.mjs: 43 pass.
node --test on write-verdict.test.mjs: 11 pass.
node --test on lane-causes-person-codes.test.mjs: 4 pass.
python3 -m unittest on test_babysit_r*.py: 708 pass, 16 skipped.
scripts/check-affected.sh --tests on the committed head.
Acceptance criteria:
A verdict on a diff with a line over 1800 bytes, every byte read: PASS in tests. Every leg's diff step on a 4000-byte line writes no stop, and its parts rejoin byte for byte to the patch. A live check needs a PR that only the main-run review binds and that touches a long line, after this merges.
A workflow.test.mjs test with a 4000-byte line: PASS.
Baseline: a PR-comment search for 2026-10-28 to 2026-11-02 found 2 PRs the merge gate held as long-line-review: (#11355, #11921). The search index may miss some.
PR: #12419 (head 372fb9e). Review-lane threads were not settled in this run: the lanes had not posted yet.
Activity
github-actions commentedon Nov 2, 2026
Proposed work class: work-class: scoped
Basis: automated intake triage; the lanes App applied this class.
Size: Medium
Complexity: Medium
Uncertainty: Known
Flags: none
Blocked by: none
Understood:
pr-review-mainstops withreason=long-lineon any diff line over 1800 bytes, so no review verdict is posted andci-statusfails. Any PR that edits one of the 371 existing over-long lines is held for a person. #12253 is an example. The issue recommends folding long lines into pieces under the cap, with a continuation marker explained in the prompt, so every byte stays readable. Done-when: a verdict thatci-statusaccepts for a diff with a long line, and a workflow test with a 4000-byte line. The issue names its own recommended approach, so I found no open intent question. I found no open issue that duplicates this one.Automated triage, applied by the lanes App.
github-actions commentedon Nov 2, 2026
github-actions commentedon Nov 2, 2026
melodic-automation-lanes commentedon Nov 2, 2026
The work-resolve-issue lane stopped on this issue (
patch-protected-path): the change needs a path the issue lane may not push, which a maintainer session builds by hand.It added
host-runand removedagent-ready: an agent run on a host builds it, and it waits on no decision from a person.In the run's words:
What the run says it did not do:
Run: https://github.com/melodic-software/claude-code-plugins/actions/runs/41387184738
To hand it back to the lane, remove
host-runand addagent-readyagain.kyle-sexton commentedon Nov 2, 2026
Correction to the Observed section: on #12253 the long-line stop did not fail
ci-status.ci-statusfailed onpr-review / claude-review-status(run 41386160944). That is the pull_request claude-review lane, and it did run: it posted a partial review that read 18 of 30 changed files. The 12 unread files were whole-file deletions, which its coverage count includes. The hold on refactor(review): remove /review:audit-enforceability #12253 (reasonlane-defect) names that cause, with a--diff-filter=dfix in ci-workflows' pr-review.long-linestop is a separate record, thepr-review-main: claude-reviewcheck run, concluded neutral withverdict=none reason=long-line. That is the main-run review. Perdocs/conventions/pr-pipeline/README.md:1259-1263, it holds a PR for a person aslong-line-review:<path>only when the PR is one "only the main-run review binds". This issue's claim that it failsci-statusfor every PR is wrong.What stands: the main-run review gives no verdict on a diff holding a line over 1800 bytes, and 371 lines in 136 tracked files on main are over it. That matters for every PR the main-run review alone binds. The folding fix in option 1 still applies to those PRs. The scope is narrower than first written.
kyle-sexton commentedon Nov 2, 2026
Claim. Backlog HITL host-run worker is taking this now.
kyle-sexton commentedon Nov 2, 2026
Host run. Built the fold that lets the main-run review read diff lines over 1800 bytes, and removed the
long-line-reviewmerge hold.What ran (Claude Code 2.1.316):
long-linestop was inline in the "Write the diff" step of each of the three legs of.github/workflows/pr-review-main.yml, and the person hold was_long_line_reasonsinplugins/source-control/skills/babysit-prs/scripts/babysit_risk.py.LONG_LINE_BYTESonto\ continued:lines, never inside a UTF-8 character. This is the same fold as pr-verify'schunk-diff.sh.write-verdict.mjsscans model text with its folds joined for credentials.max_line_bytescap is unchanged.node --testonworkflow.test.mjs: 43 pass.node --testonwrite-verdict.test.mjs: 11 pass.node --testonlane-causes-person-codes.test.mjs: 4 pass.python3 -m unittestontest_babysit_r*.py: 708 pass, 16 skipped.scripts/check-affected.sh --testson the committed head.Acceptance criteria:
workflow.test.mjstest with a 4000-byte line: PASS.Baseline: a PR-comment search for 2026-10-28 to 2026-11-02 found 2 PRs the merge gate held as
long-line-review:(#11355, #11921). The search index may miss some.PR: #12419 (head 372fb9e). Review-lane threads were not settled in this run: the lanes had not posted yet.
Target labels: add none; remove host-run.