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.
The body is a well-structured, actionable behavioral rule with concrete commands, a clear four-step workflow, validation checkpoints, and properly separated bundle files. Its only weaknesses are mild repetition of the core rule and trigger lists that duplicate the frontmatter.
Suggestions
Trim the repeated 'do NOT write from memory' admonitions to a single statement to tighten conciseness.
Spell out the Context7 invocation (tool name plus example query) instead of the 'resolve-library-id → query-docs' shorthand.
Add one line to step 4 telling the agent to loop back to step 2's doc comparison if the build or live test fails.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly lean with non-obvious, high-value advice ('A 200 with a valid-looking response body can still mean nothing happened'), but the core rule is repeated ('fetch the real documentation first', 'Do NOT write integration code from memory. Do NOT diagnose failures from memory') and 'When This Triggers' largely duplicates the frontmatter description. | 4 / 5 |
Actionability | Provides concrete executable commands (cat node_modules/<pkg>/package.json | grep '"version"'; python3 scripts/check_versions.py npm resend) and a specific code-vs-docs comparison checklist. Not 5 because 'resolve-library-id → query-docs' is abbreviated tool shorthand rather than an executable invocation. | 4 / 5 |
Workflow Clarity | Four mandatory steps are clearly sequenced with validation checkpoints (compare code vs docs, build to verify types, test live and verify the downstream effect) and the decision flowchart gives error-recovery branches. Not 5 because there is no explicit loop-back instruction when post-change verification fails. | 4 / 5 |
Progressive Disclosure | SKILL.md serves as a clear overview while bulk material is correctly split into references/doc-urls.md and scripts/check_versions.py — both real files, one level deep, each introduced with what it does and when to consult it. Navigation is easy and the split is appropriate. | 5 / 5 |
Total | 17 / 20 Passed |