Content
85%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is an exceptionally actionable and well-structured review workflow: concrete commands, explicit validation checkpoints, feedback loops on stale verdicts, and clean one-level conditional disclosure into a real bundle. Its one weakness is token efficiency — authorization and report-only rules are stated three times and several Hard Stops bullets are long run-ons that could be compressed into the existing table style.
Suggestions
Consolidate the authorization and report-only rules currently restated in the Outcome Contract, Worktree Safety Preflight, and Autofix Routing sections into a single canonical section referenced from the others.
Break the longest run-on policy sentences (e.g., the Hard Stops 'Source and distribution out of sync' bullet and the Authorization bullet) into short table rows or checklists matching the Gotchas/Autofix table style.
Deduplicate the completion-ledger and sign-off requirements stated in Outcome Contract, Specialist Review, and Sign-off into one canonical checklist, keeping the sign-off template as the single reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and free of background padding, but at ~230 lines it restates authorization and report-only rules across Outcome Contract ("may not edit files, apply autofixes, commit, push, publish"), Worktree Safety Preflight ("must not move, hide, overwrite, clean, or discard"), and Autofix Routing ("In report-only mode, do not modify the worktree"), and many multi-clause sentences could be tightened. Not 2 because nothing explains concepts Claude already knows and no section is padded; not 4 because the repetition and run-on policy sentences are more than minor. | 3 / 5 |
Actionability | Copy-paste-ready commands appear throughout ("git status --short --branch -uall", "gh run view <id> --json status,conclusion", "bash <skill-base-dir>/scripts/run-tests.sh", "git worktree add --detach <known-good-commit>"), backed by concrete decision tables (mode picker, depth, autofix routing) and a fill-in sign-off template. As an instruction-only skill, the guidance is fully executable. | 5 / 5 |
Workflow Clarity | The pipeline is clearly sequenced (preflight → mode pick → context extraction → diff → scope → review passes → verification → sign-off) with explicit validation checkpoints and feedback loops: "the old verdict expires: re-read HEAD, status, and the full resolved diff", and "Re-freeze the baseline, re-run the check that exposed the finding". The verification section also defines failure diagnosis and evidence limits. | 5 / 5 |
Progressive Disclosure | All ten references/ files and the scripts are directly linked from the body with explicit conditional load signals ("Load a mode file only when its row matches", "load references/review-patterns.md" only for matching diff shapes), each verified present in the bundle and one level deep. The SKILL.md stays an overview while mode, pattern, and catalog detail lives in the bundle. | 5 / 5 |
Total | 18 / 20 Passed |