Content
78%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.
A disciplined, token-efficient orchestration skill with a clear sequenced workflow and strong safety gating around destructive commands. Its main weakness is actionability: the pivotal step ('run the flight plan', 'smallest focused checks') references project artifacts without defining or locating them, and there is no guidance for handling failed checks.
Suggestions
Define or point to 'the flight plan' (e.g., a repo path like docs/flight-plan.md or an inline checklist) so step 2's core action is executable without guessing.
Enumerate the candidate 'focused checks' or reference where they are defined, mirroring how the proof types (UI, clinical, RAG, migration, privacy, deployment) are already listed.
Add a brief failure-handling loop: what to do when a check fails (fix and re-run vs. report and mark not-ready) before returning the ready/not-ready decision.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Fifteen lean lines with zero concept explanation — every sentence instructs ('Inspect intended diff...', 'Do not run verify:release... without approval') and the AGENTS.md boundary note is stated in one sentence. Fully matches the 'every token earns its place' anchor. | 5 / 5 |
Actionability | Concrete anchors exist (`npm run verify:pr-local`, `verify:release`, `AGENTS.md`, named artifacts like review ledger and migrations), but the core execution step — 'Run the flight plan and complete the smallest focused checks' — leaves both the flight plan and the candidate checks undefined, so Claude cannot execute step 2 without guessing. | 3 / 5 |
Workflow Clarity | A clear 6-step sequence (inspect → checks → conditional proofs → governance → gated commands → decision) with an approval gate on destructive commands and a decision output enumerating passed/failed/skipped/gated. Below the 5 anchor because there is no failure-handling loop (what to do when a check fails before deciding). | 4 / 5 |
Progressive Disclosure | Under 50 lines, single-purpose, no bundle files exist, and the content is well organized (header, boundary note, numbered list) with the one external pointer (`AGENTS.md`) clearly signaled inline. Meets the simple-skill exception for a full score. | 5 / 5 |
Total | 17 / 20 Passed |