The body is lean tables, bullets, and a decision matrix with no explanation of concepts Claude already knows (it even references "SDD forecasts" without defining them); every section earns its place, matching the score-3 anchor.

It gives concrete thresholds (400 lines, ≤60 min, 📍 marker) and a condition-to-action table, but the actual branch/retarget commands live only in the reference, so the body is not copy-paste ready as the score-3 anchor requires; it clears the score-2 bar of concrete-but-incomplete guidance.

The six-step Execution Steps sequence includes an explicit verification checkpoint ("CI/tests/docs/manual checks, rollback scope, and clean diff") and a feedback loop ("retarget or rebase until only the current work unit appears"), matching the score-3 anchor with validation present so no cap applies.

The overview is split into well-organized sections and points to a single, clearly-signaled one-level-deep reference (references/chaining-details.md, verified present) with a description of its contents, matching the score-3 anchor.

"Split oversized changes into chained PRs that protect review focus" names a single concrete action rather than the multiple distinct actions the score-3 anchor requires; it is more comprehensive than vague score-1 phrasing but not multi-action.

It explicitly answers both what ("Split oversized changes into chained PRs...") and when ("Trigger: PRs over 400 lines, stacked PRs, review slices"), with an explicit "Trigger:" clause equivalent to "Use when...", so it is not capped at 2.

"PRs over 400 lines, stacked PRs, review slices" plus "chained PRs" gives good coverage of natural terms users would say, matching the score-3 anchor; it is well beyond the single-keyword score-2 example.

The trigger set (oversized PRs, stacked/chained PRs, review slices) carves a clear niche in PR size/review-load management that is unlikely to fire for unrelated skills.