Content
96%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 orchestration skill: phased workflow with explicit exit criteria, pervasive validation checkpoints, disk-backed state management, and real helper scripts. The main improvement opportunities are de-duplicating rules repeated across Sections 1/3/4/9 and pushing the library-discovery and token-architecture detail out of SKILL.md into the already-existing phase references.
Suggestions
De-duplicate Section 9 (Per-Phase Anti-Patterns) against Sections 3–4, or collapse it into short per-phase pointers, reclaiming ~40 lines of repeated rules.
Move Section 5's get_libraries pagination behavior and reuse decision matrix into discovery-phase.md, keeping only the 'call get_libraries before search_design_system' rule and the reuse/rebuild/wrap summary in SKILL.md.
Reduce Section 8's token-architecture tables to the complexity-tier summary and defer the collection/mode examples to token-creation.md, which already covers them.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense directive content with no explanations of concepts Claude already knows — API gotchas, naming rules, and phase contracts are all skill-specific policy. It falls short of the level-5 'every token earns its place' anchor because there is real redundancy: Section 9's anti-patterns largely restate rules from Sections 3 and 4 (e.g., ALL_SCOPES, variant positioning, parallelized calls, name-prefix cleanup each appear twice), and Section 1's communication contract reiterates phase-continuation rules stated again in Section 2. It is above level 3 because the tightening opportunities are duplication of genuinely useful rules, not unnecessary explanation or concept padding. | 4 / 5 |
Actionability | Guidance is fully executable: exact tool-call signatures with parameters ('get_libraries({ fileKey })', 'search_design_system({ query, fileKey, includeLibraryKeys: [...] })'), copy-paste API patterns ('{ type: \'VARIABLE_ALIAS\', id: primitiveVar.id }', 'var(--color-bg-primary)'), a concrete state-ledger JSON shape with a disk path ('/tmp/design-system-state-{RUN_ID}.json'), and a table of eight real helper scripts with their return contracts. For an instruction-orchestration skill this is copy-paste-ready guidance covering the common cases, matching the level-5 anchor. | 5 / 5 |
Workflow Clarity | The five phases are strictly ordered with per-phase checklists, stable task IDs (P{phase}.{step}), and explicit exit criteria; validation is built in everywhere ('get_metadata after every create, get_screenshot after each component', pendingValidations in the state ledger). Error recovery has a dedicated required-on-error reference, plus a resume protocol and continuation prompt — the feedback loops and checklists the level-5 anchor requires. | 5 / 5 |
Progressive Disclosure | Structure is good and all referenced paths resolve: the Reference Docs table gives phase, required/optional status, and load conditions for all seven reference files, and the Scripts table maps all eight scripts. It misses the level-5 anchor because substantial detail is inlined in SKILL.md that belongs in those references — Section 5's ~50 lines of get_libraries pagination behavior and reuse matrix belong in discovery-phase.md, and Section 8's token architecture overlaps token-creation.md (which itself notes the M3 pattern). All bundle references were verified to exist on disk; one cross-skill link (../figma-use/references/gotchas.md) points outside this bundle, which is expected for a companion-skill link. | 4 / 5 |
Total | 18 / 20 Passed |