Content
92%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 high-quality instruction-only skill body: every scenario is backed by executable commands, exact manifest requirements, and failure signatures with specific recoveries, and risky operations are guarded by validation and confirmation gates. The only weakness is mild repetition (the 2-day expiry window appears three times, and the example dialogue restates the new-package steps).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes Claude's competence — no concept explanations, just operational facts like "Publishing is sequential, so the packages after it were skipped, not broken". However, the 2-day trust-expiry window is stated three times (step 4, the example dialogue, and the workflow-change section) and the example block partially restates steps 1–4, so minor trimming is possible. Fits the 4 anchor ("efficient; minor instances that could be trimmed") rather than 5 ("every token earns its place"). | 4 / 5 |
Actionability | Fully executable, copy-paste-ready commands throughout: `pnpm npm-trust preflight`, `pnpm npm-trust add <name>`, `pnpm npm-trust verify --package <name>`, `gh workflow run release.yml --ref main -f tag=<tag>`, `gh run view <run-id> --log-failed`, exact manifest JSON fields, and the tag regex `^[a-z][a-z0-9-]*$`. Failure classification maps each error signature to a specific recovery action. Matches the 5 anchor ("copy-paste ready code or commands; specific examples cover the common cases"). | 5 / 5 |
Workflow Clarity | Each scenario is a clearly sequenced workflow with explicit validation checkpoints and feedback loops: new-package flow runs preflight (expected "not on npm yet" failure documented), mandates the maintainer action with timing, and verifies after release; the diagnosis flow identifies the first failing package, classifies the cause, gives a per-class recovery command, and re-runs verify before concluding ("Re-run `verify` before concluding that something is missing"). Registry-mutating commands are gated behind explicit user confirmation, so the destructive-operation validation cap does not apply. Matches the 5 anchor (explicit validation steps; feedback loops for error recovery). | 5 / 5 |
Progressive Disclosure | With no bundle files, the single SKILL.md acts as a well-sectioned overview that delegates depth one level deep with clearly signaled references: "The full runbook is `contributing/npm-publishing.md`; the decision record is `docs/adr/0002-npm-trusted-publishing.md`. Read the runbook section that matches the request instead of restating it", plus targeted pointers for changesets and monorepo structure. Sections (When to use / not, Required inputs, per-scenario Instructions, Expected output, Safety) make navigation trivial. Matches the 5 anchor (clear overview, well-signaled one-level-deep references) rather than 4 because no inlined content belongs in a separate file. | 5 / 5 |
Total | 19 / 20 Passed |