Content
75%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 tight, well-sequenced multi-phase workflow with concrete executable commands and explicit validation checkpoints throughout. The main gaps are a lack of a pre-check before the destructive Phase 4 cleanup and a few indirect references to the sibling /understand skill.
Suggestions
Add an explicit validation or safety check before the Phase 4 `rm -rf` cleanup (e.g., confirm $INTER resolves under $UA_DIR/intermediate before deleting) to strengthen the destructive-operation checkpoint.
Inline or briefly summarize the "/understand directive text" and "Phase 0.1.5" logic referenced indirectly, so the skill is self-contained rather than depending on a sibling skill's body.
Provide a minimal executable example for the Phase 2 ANALYZE batch dispatch (e.g., a sample agent invocation command) to lift actionability from mostly-executable to copy-paste-ready.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a lean, phase-based workflow that assumes Claude's competence (no padding about what Figma or a knowledge graph is); every phase carries task-relevant commands and notes. Not 5 because the inline $UA_DIR resolution explanation ("Because each phase may run in a fresh shell, carry $UA_DIR forward...") is dense and could be trimmed; not 3 because it is efficient overall with only minor over-explanation. | 4 / 5 |
Actionability | Provides concrete executable commands such as `node <SKILL_DIR>/figma-scan.mjs "$PROJECT_ROOT" "<url-or-key>"`, the pnpm build, mkdir, and the find cleanup, with placeholders defined in context. Not 5 because Phase 2 (ANALYZE) is procedural guidance ("Group nodes into batches of ~15", "dispatch a subagent") rather than copy-paste-ready code; not 3 because the provided commands are real and executable, not pseudocode. | 4 / 5 |
Workflow Clarity | Clear five-phase sequence (Phase 0-4) with explicit checkpoints: Phase 1 STOP on non-zero exit and on UP_TO_DATE, Phase 0 build-or-fail, Phase 3 merge that "validates", and Phase 2 batch error handling ("If a batch fails, log a warning and continue"). Not 5 because the destructive `find ... -exec rm -rf {} +` cleanup in Phase 4 lacks an explicit pre-validation step; not 3 because validation is present across the workflow rather than missing, so the destructive/batch cap does not strictly apply. | 4 / 5 |
Progressive Disclosure | Well-organized into Prerequisites and Phases 0-4 with a one-line overview ("Analyzes a Figma file and produces an interactive design knowledge graph in the existing dashboard") and clearly signaled one-level-deep references (`<SKILL_DIR>/figma-scan.mjs`, `agents/design-analyzer.md`, the `/understand-dashboard` skill). Not 5 because some references are indirect (reusing "/understand's directive text" and "Phase 0.1.5" logic) and no bundle files are present to verify the referenced scripts; not 3 because structure and signaling are good, not merely present. | 4 / 5 |
Total | 16 / 20 Passed |