Content
71%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.
Extremely actionable, repo-specific content with genuine validation gates — the strongest part of the skill. The weaknesses are structural: a ~1050-line monolithic body with no bundle files, heavy reference material inlined where progressive disclosure expects separate reference files, and a closing checklist that duplicates the body's content. Splitting the reference sections into references/ files and trimming the redundant checklist would raise both conciseness and progressive disclosure without losing guidance.
Suggestions
Move the SubBlock Types Reference, File Input Handling, and OAuth deployment-availability sections into references/ files (e.g. references/subblock-types.md, references/file-input.md) and keep one-line pointers in SKILL.md — this fixes both the monolithic structure and the token budget.
Replace the 29-item checklist with a short pointer to the 'Final Validation' section, or deduplicate it against the body so each gate appears once; nearly every item restates a rule already stated above.
Expand the opening 'Your Task' into the real ordered sequence (author block and meta → register in registry-maps.ts → regenerate generated artifacts → run checks) so the workflow does not have to be reconstructed from scattered sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~1050-line body is dense and mostly genuinely non-obvious repo knowledge (canonicalParamId serialization semantics, deployment gating, generated-artifact rules) rather than concepts Claude already knows, so it does not fall to score 2. But it is noticeably over budget for a SKILL.md body: the SubBlock Types Reference, File Input Handling, and Generated Artifacts sections are full reference manuals inlined rather than pointed to, and the closing checklist re-states body content almost line for line. Mostly efficient content, delivered in a form that could be tightened considerably — anchor 3. | 3 / 5 |
Actionability | Nearly every section gives executable TypeScript, exact file paths, exact function names (getScopesForService, createVersionedToolSelector, normalizeFileInput), and runnable commands (`bun run docs:check`, `bun run check-canvas-sentences.ts --block={service}`). The complete annotated BlockConfig example plus the explicitly good/bad snippets make the guidance copy-paste ready and cover common cases. | 5 / 5 |
Workflow Clarity | The 'Final Validation (Required)' section gives a numbered sequence with explicit verification of each tool in tools.access, the checklist enumerates every gate, and validation commands with failure feedback loops are given throughout (docs:check failure on stale artifacts, check-block-registry rules, fork-dependent-coverage). Not score 5 because the opening 'Your Task' sequence is a three-line stub and the real ordering of work (author, register, regenerate artifacts, validate) must be inferred by stitching together sections scattered across the document. | 4 / 5 |
Progressive Disclosure | No references/, scripts/, or assets/ bundle files exist — everything lives in one monolithic SKILL.md, and ~600 lines of it (SubBlock Types Reference, File Input Handling, OAuth deployment availability, Generated artifacts) is exactly the reference material that anchor 3/2 says belongs in separate, on-demand files. It scores 3 rather than 2 because the body is well-sectioned with clear headers and the pointers it does give are clearly signaled and one level deep (apps/sim/blocks/AGENTS.md, the /add-trigger and add-selector skills, .agents/skills/tool-registry-boundary/SKILL.md). | 3 / 5 |
Total | 15 / 20 Passed |