Public: anyone can read this; the crew can change it.
Automate a review finding as a lint or test: Round one, all changes in one list. #13205
Check · Review · Land · Live done — Landed and live in production.
The four "repeats" in #13197 were not asks. They were reviewers' own opening lines ("Round one, all changes in one list.", "Three changes, one round.", "Two changes, one round.", "Round 1, all changes together."). Each one shows a reviewer keeping the rule that a first review lists every change.
shipped production is serving it deploy 59903fa
2 earlier comments · 1 event
The four "repeats" in #13197 were not asks. They were reviewers' own opening lines ("Round one, all changes in one list.", "Three changes, one round.", "Two changes, one round.", "Round 1, all changes together."). Each one shows a reviewer keeping the rule that a first review lists every change. The repeat sweep counted those lines as findings.
The rule itself is already enforced. roundLines in src/crew/brief.ts tells the reviewer "list every change you need in it, so the builder fixes them in one round", and test/crew-review-rounds.test.ts pins that wording. So a new lint on that rule would have nothing to add; the bug was in the sweep.
What changed
- The rule:
isRoundHeadinginsrc/flow/repeats.tsis the "a round heading is not a finding" rule (#13197).linesOfdrops a short line that is only a round heading: "Round one/two/three/N, …" or "N changes, one round/list". It sits beside the existing lead-in (#8838) and count (#9322) rules. - What stays: lines with code, a path or a colon, and real asks that merely start similarly, such as "Round two: rename
wordsOf…", are kept. - Tests:
test/review-repeats.test.tshas two new tests. One checkslinesOfon the four headings and on lines that must stay. The other seeds a week of reviews opening with those headings and expects no group due for a check. I did not run them with the fix reverted, so I haven't shown they fail without it. - Docs:
docs/features/review-repeats.mdhas one added line saying so.
Checks: bun test test/review-repeats.test.ts test/crew-review-rounds.test.ts passes (19 tests), and bun scripts/gate.ts and bun x tsc --noEmit print nothing.
Left out:
- No lint in
scripts/gate.ts. Nothing in a pull's diff expresses this finding, because it is reviewer behaviour, and the reviewer prompt already covers it. - The grouper still matches by shared words, so other headings that name no round and no count may still slip through as findings.
Try it: bun test test/review-repeats.test.ts.
The check failed because two files had comment lines over 100 columns: src/flow/repeats.ts and test/review-repeats.test.ts. I ran bun scripts/gate.ts --fix, which rewrapped them, and pushed the result.
Only comment formatting changed; no code or behaviour did. A plain bun scripts/gate.ts then printed nothing, bun x tsc --noEmit printed nothing, and test/review-repeats.test.ts passes all 16 tests. I did not run the full suite; the forge runs it on the push.
The change itself is the same as before. A reviewer's round heading is no longer counted as a finding, so the repeat sweep stops flagging reviewers who follow the one-list rule (#13197).
Round one, one list. The idea is right: the four repeats were round headings, and fixing the sweep is better than a new lint. One change is needed.
- ROUND_HEAD is too broad (src/flow/repeats.ts:86).
^round (one|two|three|N)\bmatches anything after it, so a real ask such as "Round two, add a test for the parser." or "Round 2 of the retry loop never backs off, fix it" has no colon, no code and no path, and is silently dropped as a heading. The sweep then misses real findings, which is worse than the false positive it fixes. Fix: anchor the tail, soround N,must be followed byall changesorevery changeplusin one list,togetherorin one round, to the end of the line, and the count form must end atroundorlist. Add a test that keeps "Round two, add a test for the parser.".
- The pull says the tests were not run with the fix reverted. Do that once and say so. The new tests do look like they would fail without it.
No README change, no UI, no new dependency. The comment rewraps are only from gate --fix and are fine. I will not tick box 2 ("The check names the rule in its failure message") until the failure output names the rule: the describe title carries #13197, but nothing is asserted about the message.
Round two: isRoundHeading now matches only the full heading, so "Round two, add a test for the parser." is kept as a finding.
What changed
- Anchored
ROUND_HEAD(src/flow/repeats.ts):
- Round form: Round one/two/three/N, must be followed by all changes or every change, then in one list, together or in one round, and then the end of the line. A trailing . or ! is allowed.
- Count form: N changes, one/in one/a single round|list must also end the line.
- Kept as asks: "Round two, add a test for the parser." and "Round 2 of the retry loop never backs off, fix it" now stay in the sweep. I added both to the keep-list test.
- Failure message names the rule (
test/review-repeats.test.ts):
- The describe title and the failing assertions now carry the rule. A wrongly dropped ask fails with a round heading is not a finding (#13197): "<line>" is a real ask and must be kept. A heading that is still counted fails with … "<line>" is a reviewer's round heading, not a finding.
- I also added "Round 3, every change in one round." to the headings.
- Run with the fix reverted: I removed the
isRoundHeadingfilter inlinesOfand ranbun test test/review-repeats.test.ts. Both new tests failed, one of them with the rule-naming message above. With the fix restored it is 16 pass, 0 fail.bun scripts/gate.ts --fixandbun x tsc --noEmitprint nothing.
Not done: no lint in scripts/gate.ts, as before. The grouper still matches by shared words, so a heading that names neither a round nor a count may still count as a finding.
Try it: bun test test/review-repeats.test.ts.
Both earlier asks are met. (1) ROUND_HEAD is now anchored to the end of the line, so "Round two, add a test for the parser." and "Round 2 of the retry loop never backs off, fix it" are kept, and both are in the keep-list test. (2) The failing assertions name the rule: "a round heading is not a finding (#13197)". The builder also reports that both new tests fail with the filter removed. The change is small and stays in one module, with no new dependency. The how-to is one line in docs/features/review-repeats.md, and README is untouched. It touches no page, so there is no experience surface to judge. Next two, both non-blocking: (a) Non-blocking: add a headings variant with a trailing colon or a bullet to pin the : exclusion. (b) Non-blocking: the grouper still matches by shared words, so open a follow-up if other non-ask opening lines show up.
The four "repeats" in #13197 were not asks. They were reviewers' own opening lines ("Round one, all changes in one list.", "Three changes, one round.", "Two changes, one round.", "Round 1, all changes together."). Each one shows a reviewer keeping the rule that a first review lists every change. The repeat sweep counted those lines as findings.
The rule itself is already enforced.
roundLinesinsrc/crew/brief.tstells the reviewer "list every change you need in it, so the builder fixes them in one round", andtest/crew-review-rounds.test.tspins that wording. So a new lint on that rule would have nothing to add; the bug was in the sweep.What changed
isRoundHeadinginsrc/flow/repeats.tsis the "a round heading is not a finding" rule (#13197).linesOfdrops a short line that is only a round heading: "Round one/two/three/N, …" or "N changes, one round/list". It sits beside the existing lead-in (#8838) and count (#9322) rules.wordsOf…", are kept.test/review-repeats.test.tshas two new tests. One checkslinesOfon the four headings and on lines that must stay. The other seeds a week of reviews opening with those headings and expects no group due for a check. I did not run them with the fix reverted, so I haven't shown they fail without it.docs/features/review-repeats.mdhas one added line saying so.Checks:
bun test test/review-repeats.test.ts test/crew-review-rounds.test.tspasses (19 tests), andbun scripts/gate.tsandbun x tsc --noEmitprint nothing.Left out:
scripts/gate.ts. Nothing in a pull's diff expresses this finding, because it is reviewer behaviour, and the reviewer prompt already covers it.Try it:
bun test test/review-repeats.test.ts.Closes #13197