Content
82%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 tight, highly actionable phased workflow with genuine validation checkpoints and no padding. The main gaps are deferred executable details (build-verification and language-directive text borrowed from /understand), missing failure handling for the merge phase before a destructive cleanup, and bundle references that cannot be verified against the packaged files.
Suggestions
Inline the core-build verification as a concrete command (or a one-line equivalent) instead of deferring to '/understand Phase 0.1.5' logic, and restate the $LANGUAGE_DIRECTIVE template rather than pointing at /understand's text.
Add explicit failure handling for Phase 3 (e.g. 'If figma-merge.mjs exits non-zero, relay stderr and STOP') and make the Phase 4 rm -rf cleanup conditional on a successful merge, since it is destructive.
Ship the referenced bundle files (figma-scan.mjs, figma-merge.mjs, agents/design-analyzer.md) alongside SKILL.md, or list them in a short 'Bundle files' section so the references are verifiable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~55-line body is lean and entirely operational: phased sections with exact commands, batch sizes, and concurrency limits, and no explanation of concepts Claude already knows. Even the dense UA_DIR paragraph earns its tokens by solving a real fresh-shell state problem, so every line adds non-obvious information. | 5 / 5 |
Actionability | Most guidance is copy-paste executable (pnpm build command, figma-scan.mjs and figma-merge.mjs invocations with arguments, the find-based cleanup, batch of ~15 with 5 concurrent), but two steps defer executable details to another skill's documentation: 'ensure core is built (same logic as /understand Phase 0.1.5)' and 'reuse /understand's directive text'. It is well above 3 because the commands are real and complete rather than pseudocode or high-level hints. | 4 / 5 |
Workflow Clarity | Phases 0–4 are clearly sequenced with real checkpoints: Phase 1 has 'If it exits non-zero, relay stderr and STOP' plus an UP_TO_DATE short-circuit, and Phase 2 tolerates batch failure explicitly. It falls short of 5 because Phase 3 (merge) has no stated failure handling, and the Phase 4 rm -rf cleanup — a destructive batch operation — has no explicit precondition that the merge succeeded. | 4 / 5 |
Progressive Disclosure | The single file is well organized into prerequisites and five phases, appropriately delegating implementation to scripts and an agent definition rather than inlining them. However, no bundle files exist in the skill directory, so the referenced paths (<SKILL_DIR>/figma-scan.mjs, figma-merge.mjs, agents/design-analyzer.md) are unverified and would be dangling as packaged, which is a minor organization gap. | 4 / 5 |
Total | 17 / 20 Passed |