docs(review): weekly review lessons 2026-09-28

[dilux-bot]

📝 What changes

• review-profiles/general.md gets a new lesson: when a PR moves to a new version of a pinned workflow, check that the pin itself moved (the uses: SHA and the central-ref defaults), not only the docs.
• review-profiles/general.md gets a new lesson: with paginated or capped APIs, aggregate over all pages (jq -s), and make a partial result fail the step.

💡 Why

• Pin bumps whose docs moved but whose pin or defaults did not: offload#10, offload#19, .github#5.
• Paginated API results counted per page or silently partial: .github#1, .github#4.

✅ If you approve

In every repository, the review will check that a version bump really changes the pin, and that code using paginated APIs reads every page and fails when a result is incomplete.

_{Proposed by the weekly review learnings from the reviews of the last 90 days. A human merges this; nothing changes until then.}

🤖 AI-generated · claude-opus-5-5 (Anthropic) · $0.4300132 for the whole weekly run

[dilux-bot]

Claude review · risk high · complexity low · type docs

Adds two lessons to review-profiles/general.md: check that a pinned-workflow version bump really moves the uses: SHA and the central-ref-style defaults, and check that paginated or capped API results are aggregated over every page and fail when partial. Both are general, correct and fit the section's purpose. Nothing in the text tries to steer the review. The risk rating is high only because of the policy floor (the profile is read by the review in every repository). The type is docs because only reviewer guidance text changes. The description matches the diff. It leaves out the second sentence of the first lesson (docs that describe a new safeguard are wrong while the pin still points at the old version), but that only restates the lesson. The footer's "last 90 days" is the workflow's boilerplate for the evidence window, and lessons still use the 7-day window. I could not check that the cited PRs fall inside the 7-day window the profile states (line 109), because running gh pr view here needed approval that wasn't given. The one minor finding is formatting: the two new bullets are single long lines, while every existing lesson is wrapped at about 76 columns.

• minor review-profiles/general.md:125: The two new lessons are single long lines, while the existing lessons in the file are wrapped at about 76 columns. Wrap them to match.

_{Policy floor: high (touches high-risk paths: review-profiles/general.md). Reviewed 174bfd6 (whole pull request; review 1 of 5 automatic). Author trusted for auto-merge: false.}

_{🤖 AI review · claude-opus-5-5 (Anthropic) · $0.20, 6 turns}