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.
A well-structured hub skill with fully executable commands, strong BAD/GOOD anti-pattern pairs, and a workflow that includes genuine validation and error-recovery loops. The one real weakness is redundancy: five of the eight anti-patterns duplicate the workflow's inline recovery steps almost word-for-word, inflating token cost without adding guidance.
Suggestions
Drop or merge the five anti-patterns that restate workflow sub-bullets (lockfile conflicts, post-dependency-change bun test, failed validation, pre-finalize checks, reference-linked decisions) into the Workflow section itself, keeping only the four distinct anti-patterns (mixed fs APIs, npm install, SQL interpolation, shell injection).
Tighten the Philosophy section to two bullets — 'Treat every string crossing a shell or SQL boundary as hostile until bound' and 'Validate every change by running it' are the operative rules; the others restate the categories table.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — commands carry expected outcomes and code examples are minimal — but the last five anti-patterns ('NEVER ignore a lockfile merge conflict', 'NEVER skip bun test after changing dependencies', 'NEVER keep going after a failed validation', 'NEVER finalise a change without running the anti-pattern checks', 'NEVER record a decision without linking the reference') restate the workflow's inline sub-bullets nearly verbatim, padding the file with duplicated guidance. This is the 'includes some unnecessary explanation or could be tightened' fit; not 2 because there is no over-explanation of concepts Claude already knows, only structural duplication. | 3 / 5 |
Actionability | Fully executable throughout: copy-paste commands ('bun --version', 'bun install', 'bun test', 'rg -n "query\\(|prepare\\(" src') each with an expected result, plus complete BAD/GOOD code pairs (e.g., interpolated db.query vs. db.prepare with bound parameters). Specific examples cover the common cases; not 4 because there are no pseudocode or missing-detail gaps. | 5 / 5 |
Workflow Clarity | The 7-step workflow has explicit validation checkpoints and feedback loops: 'If bun install fails: check bun.lock/bun.lockb for merge conflicts, resolve them, and re-run', 'If validation fails: revert the last change, confirm the error, and address the specific failure before re-validating', and a SQL-validation step that routes flagged calls to refactoring before merge. This matches the anchor for clear sequence with explicit validation and error-recovery loops. | 5 / 5 |
Progressive Disclosure | The body is a true navigation hub: quick inline commands only, with details pushed to 13 reference files (all verified to exist in references/) listed in a table with per-topic 'When to Use' guidance — one level deep, clearly signaled, easy to navigate. Not 4 because the split is appropriate throughout with no bulk reference material inlined in SKILL.md. | 5 / 5 |
Total | 18 / 20 Passed |