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.
An exemplary procedural skill: concrete, sequenced, validated, and well-factored across its bundle. The only noteworthy defect is a frontmatter/body inconsistency — the description claims the skill 'bumps the CLAUDE.md tool count' while step 8 explicitly says 'Do not bump any tool count — docs/MCP_TOOLS.md is generated and authoritative' — which should be reconciled to avoid misleading execution.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost entirely repo-specific procedure (collision greps, annotation lines, e2e/tools.ts updates) rather than concepts Claude already knows, but a few rationale passages run longer than needed — step 3's multi-sentence read/write justification ('Getting it wrong has runbook consequences... so it's never correct') and step 9's closing paragraph ('Better to handle them deliberately than rush at the tail of adding a tool') could be tightened. This matches 'efficient; minor instances of over-explanation that could be trimmed' rather than the lean-everywhere anchor at 5. | 4 / 5 |
Actionability | Fully executable guidance throughout: exact commands ('Grep src/<provider>/server.ts for name: \'<toolName>\'', 'npm run typecheck'), enumerated template placeholders ({{toolName}}, {{zodFields}}, {{executeBody}}), exact annotation lines ('annotations: { readOnlyHint: false, destructiveHint: true },'), a copy-paste report block, and verbatim follow-up offer text. Code generation is anchored to real template files in assets/templates/. | 5 / 5 |
Workflow Clarity | Eleven clearly sequenced steps with explicit validation checkpoints and a feedback loop: step 1 confirms the target file exists, step 2 aborts on name collision, and step 10 runs typecheck with enumerated failure modes and fixes ('Zod schema doesn't match what the execute body destructures'). This matches the anchor with explicit validation steps and error-recovery guidance; checkpoints are explicit, not merely present. | 5 / 5 |
Progressive Disclosure | Clean split confirmed against the actual bundle: SKILL.md is the insertion procedure, the canonical pattern lives in references/tool-pattern.md ('Read references/tool-pattern.md before generating'), and code lives in assets/templates/google-tool.ts.tmpl and third-party-tool.ts.tmpl — all referenced paths exist, and tool-pattern.md itself contains no nested references, so references are exactly one level deep. The File layout section documents the navigation. | 5 / 5 |
Total | 19 / 20 Passed |