docs(review): state the nit cap in its own per-concern review file #12198

Refs: #7583

Why: #7583's Decision "the nit cap lives in its per-concern file (Q570)" is the one decision of that issue not yet delivered. No file under docs/review/ or REVIEW.md caps nits today. The only nit text is the severity row | Suggestion | 🟡 Nit | at REVIEW.md:31.

Summary

Add a per-concern review file that limits how many Suggestion-severity findings (🟡 Nit in REVIEW.md's severity table) a single review posts. Then index it in docs/review/README.md with a trigger that fires on every review. REVIEW.md is synced from melodic-software/standards and stays unchanged. Its "Depth" section already points self-hosted and local reviews at docs/review/README.md.

Fix

1. Create docs/review/findings.md with a # Findings heading and this rule as one list line, in the shape of the other files in docs/review/ (rule text, then a link):
   • A review posts at most five Suggestion-severity findings (🟡 Nit), choosing the five with the most consequence. It reports how many more it found as a count in the summary, and posts none of them inline. Critical and Important findings, blocking findings, and manifest <field>: flag lines are never capped: Code Review: Choose what to write in REVIEW.md.
2. Add one row to the table in docs/review/README.md, directly under the REVIEW.md row:
   | [`findings.md`](findings.md) | anything: how many findings one review posts |
3. Leave REVIEW.md unchanged. A change there is melodic-software/standards work, which standards: per-concern standards layout #7583 marks as an owner stop.

Basis:

• The cap moves to a per-concern file: owner answer Q570 (poteto-review interview ledger) and standards: per-concern standards layout #7583's Decisions.
• The number five comes from the upstream example, not from an owner decision. https://code.claude.com/docs/en/code-review-customize (read 2026-11-01) has "Report at most five Nits per review." under the "Cap the nits" heading of its example REVIEW.md, and, under "Choose what to write in REVIEW.md", says a cap of five nits "keeps reviews actionable".
• The excess is reported as a count: judgment. The upstream page does not say what happens to nits past the cap; the poteto-review research draft recommends "report nits as a count" for incremental reviews, and a count keeps the dropped volume visible without posting it.
• A new file and not pull-requests.md: judgment. pull-requests.md covers author-side post-PR triage, and the index loads a file only when the change touches its concern, so a cap there would not load on most reviews.
• The CI code-review lane (plugins/review/skills/code-review/SKILL.md, "High-signal bar") already excludes style nits; this rule does not change that skill.

Done when

• docs/review/findings.md exists, and grep -n "at most five Suggestion" docs/review/findings.md prints one line.
• grep -n "count" docs/review/findings.md prints the line saying the findings past the cap are reported as a count in the summary.
• grep -n "never capped" docs/review/findings.md prints the line exempting Critical, Important, blocking and manifest <field>: flag findings.
• grep -n "code-review-customize" docs/review/findings.md prints the rule line's upstream link.
• grep -n "(findings.md)" docs/review/README.md prints one table row whose second column starts with anything.
• git diff --quiet origin/main -- REVIEW.md exits 0.
• grep -cP '\x{2014}' docs/review/findings.md docs/review/README.md prints a count of 0 for each file (no em dash).
• git diff --name-only origin/main lists only docs/review/findings.md and docs/review/README.md.

Verification

Run the grep and git commands in Done when against the PR head.

Related

• standards: per-concern standards layout #7583 (parent decision), standards: per-concern standards layout - part 1 of 2 #10231 (part 1: AGENTS.md review rules moved into docs/review/, closed by docs(review): move the AGENTS.md code review rules into per-concern docs review files #12124), standards: per-concern standards layout - part 2 of 2 #10232 (part 2: the REVIEW.md generation drift check, open).
• Out of scope: the nit cap for Managed Code Review. It honors a nit cap only from the root REVIEW.md or a file beside it that REVIEW.md includes, and an include of a path under docs/ is not followed (https://code.claude.com/docs/en/code-review-customize, "Include other files in REVIEW.md"). That would be melodic-software/standards work.