Content
100%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 skill body: a strict spec-to-render pipeline with exact commands, validation and error-recovery checkpoints at every risky step, terse house rules expressed only as project-specific doctrine, and a fully verified one-level-deep reference bundle. The Red Flags table and anti-patterns add actionable guardrails without wasting tokens.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean throughout: "Never hand-write mxGraph XML. Write a spec; the scripts own every visual decision" and the one-line rules ("One C4 level per diagram", "Exec audience caps at 12 nodes") assume Claude's competence and add only project-specific doctrine. No concepts Claude already knows (what C4 or draw.io is) are explained, and the Red Flags table is compressed thought-correction, not padding — matching the lean/every-token-earns-its-place anchor at 5 rather than the minor-trimming level at 4. | 5 / 5 |
Actionability | Every pipeline step is a copy-paste-ready command with flags and placeholders: `python3 scripts/schema_to_spec.py db/schema.sql --title "<System> — ERD" -o spec.json`, `python3 scripts/validate_spec.py spec.json`, `python3 scripts/render_drawio.py spec.json -o docs/architecture/<slug>.drawio --strict`, plus the export command with an explicit fallback chain and exit-code semantics ("exit 2 = a layout finding"). Guidelines use concrete field names (`metric`, `constraint`, `style: async`, `gcp:*`, `aws:*`, `cloud:*`, 12-node cap), fully matching the copy-paste-ready anchor at 5. | 5 / 5 |
Workflow Clarity | The 6-step pipeline is clearly sequenced with explicit validation checkpoints and feedback loops: step 2 validates the spec, step 3 explains error recovery by exit code ("change the spec, per layout-rules.md"), and step 6 closes the loop ("Inspect the exported image... Fix the spec and re-export before handoff") with an honest no-tool fallback ("ship the .drawio and say the image was not exported"). A checklist reference exists for complex runs, satisfying the anchor-5 criteria of validation steps, error-recovery loops, and checklists rather than the minor-gaps level at 4. | 5 / 5 |
Progressive Disclosure | SKILL.md is a concise overview with well-signaled, one-level-deep references: all 12 `references/*.md` files, the 5 pipeline scripts, and `assets/fixtures/<type>.spec.json` (verified present, one per each of the 8 diagram types, plus schema samples) exist in the bundle, and reference files link only to siblings — no 2+ level nesting. Content is appropriately split (spec schema, selection guidance, style catalog, layout rules, export paths each in their own file) with a consolidated References section for navigation, exactly matching the anchor-5 example. | 5 / 5 |
Total | 20 / 20 Passed |