ums(2071): gha SemBr check enforces rules the local script does not

[dem-extra1]

Closes nothing; banks the UMS learning from #2071. Follow-up issue for the tools.md split: #2093.

One new section in memories/github-actions.md: the repo-local scripts/semantic-line-breaks.py and the gha check-new-line-breaks.yml enforce different rule sets (gha additionally flags long lines with mid-line semicolons), so a green local run does not predict the CI check. Pattern/anti-pattern pair included per corpus style.

[dem-extra1]

All three findings addressed: reverted the memories/github-actions.md fork (which removes the violating line, the pronoun, and the duplication in one stroke) and appended the #2071 instance to shared/writing/semantic-line-breaks.md as a fourth dated recurrence cross-referencing #2085, per challenge-redundant-content.

[github-actions]

Claude finished review — View run

---

Confirmed open, unmerged, not draft. Review complete.

Code review — #2095

Summary: This PR has been substantially revised since the last review round. The original commit (4f55705f9, reviewed previously) added a section to memories/github-actions.md that duplicated existing content and contained a CI-breaking mid-line-semicolon line. The author reverted that commit and replaced it with a new commit (1555707) that instead appends a "Fourth dated recurrence" section directly to shared/writing/semantic-line-breaks.md — exactly the consolidation the prior review suggested. I verified:

• Prior finding 1 (CI-breaking line): resolved. new-line-breaks / check-new-line-breaks now passes on this head; I confirmed no line in the new block exceeds 80 chars or contains a semicolon.
• Prior finding 2 (redundancy with existing fragment/issue): resolved. Content now lives in the fragment itself as a new dated-recurrence entry, following the file's established convention (a "Third dated recurrence" entry already exists at line 122), rather than forking a disconnected record.
• Prior finding 3 (ambiguous pronoun): moot. The memories/github-actions.md text it applied to was reverted.
• Technical claims verified against source: the referenced commit pair 456a6c87 → 571160436 (both fetched and diffed) shows exactly the described fix — two clauses wrapped at their semicolons and one parenthetical reworded to drop its semicolon, three violations total, matching the text precisely. Issue scripts/semantic-line-breaks.py and CI's check-new-line-breaks disagree on mid-line semicolons #2085 and PR feat(scripts): pairwise PR file-set collision instrument (closes #2072) #2073 both exist and match the described root cause.

New finding

1. (Non-blocking, but a genuine factual error) "hours earlier" misstates the actual time gap, which was minutes.
shared/writing/semantic-line-breaks.md line 171 (in the new block, ~L164-184): "[Issue #2085]... records the same root cause from PR #2073 hours earlier."

I checked the actual timestamps:

• PR memories: record MCP tool search default-on #2071's first failing commit (456a6c87): 2026-08-24T05:50:51Z
• PR feat(scripts): pairwise PR file-set collision instrument (closes #2072) #2073's CI failure documented in issue scripts/semantic-line-breaks.py and CI's check-new-line-breaks disagree on mid-line semicolons #2085 (job 97335526930): 2026-08-24T05:52:19Z–05:55:12Z
• Issue scripts/semantic-line-breaks.py and CI's check-new-line-breaks disagree on mid-line semicolons #2085 filed: 2026-08-24T05:55:33Z
• PR memories: record MCP tool search default-on #2071's fix commit (571160436): 2026-08-24T06:00:16Z

Every one of these events falls within a ~10-minute window (05:50Z–06:00Z), and even against the widest reasonable reference point (PR #2071's own creation at 05:13Z), the gap is under an hour — never "hours." Given this file's own stated purpose is precise, checkable dated-recurrence record-keeping (per shared/writing/timestamp-volatile-claims.md's "attach the time the claim was true" and shared/writing/fact-check-prose.md), this is worth correcting to "minutes earlier" (or simply dropping the qualifier) before merge. It's not blocking — the surrounding factual claims (commit SHAs, the three-semicolon count, the "same day" framing) all check out — but it's a specific, verifiable inaccuracy in a passage whose entire value is being precisely dated.

Verdict

Ready for merge — the two blocking issues from the prior review round (CI-breaking line, redundant content) are both resolved through the author's revision, and the technical claims I independently checked (commit diffs, issue/PR cross-references, semicolon count) hold up. The one new finding ("hours earlier" vs. the actual ~10-minute gap) is a minor factual imprecision, not a blocker; recommend fixing it in this PR or a quick follow-up, but it shouldn't hold up the merge.

Reviewed commit: 1555707

[github-actions]

💰 Cost: $2.3674 (review) — run