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.
A lean, highly actionable skill body with an exemplary gated workflow: reproduction before change, explicit decision criteria, and per-issue verification evidence. The only weakness is bundle integrity — the well-signaled reference/ and harness/ paths referenced throughout do not exist in the provided bundle, which keeps progressive disclosure at 4.
Suggestions
Ship the referenced files: the body's References section and Tools table point to .agents/skills/design-clip-repro/reference/*.md and harness/* (doctor.mjs, dlib.mjs, sheet.mjs, figlib.mjs, lock.mjs, shot.mjs, osmouse), none of which are present in the bundle — include them so the one-level-deep navigation actually resolves.
Add export-figma-cookies.mjs to the body (e.g., in the Figma login row as the way to produce/refresh the FIGMA_COOKIES_B64 secret), since it is the only bundled script never mentioned; alternatively remove it from the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and imperative throughout — no concept Claude already knows is explained, no padding. Every line carries operational information the model could not invent: the Fusion/machine table, exact helper signatures ("openEditor(id, { prod, reuse })"), the OOM warning with "NODE_OPTIONS=--max-old-space-size=4096", and judgment rules like "Never prove a bug on a design you built yourself." The one explanatory passage (Figma text corner-radius rationale) is repo-specific knowledge that justifies its tokens, so anchor 5's 'every token earns its place' fits rather than anchor 4's 'minor instances of over-explanation'. | 5 / 5 |
Actionability | Fully executable guidance: exact commands ("node .agents/skills/design-clip-repro/harness/doctor.mjs", "scripts/clip.mjs \"<link>\" [atMs,...]", "swiftc -O ...osmouse.swift -o .../osmouse"), a complete API table with function signatures and return values, and where artifacts go ("templates/design/.tmp/parity/shots/"). Copy-paste ready and covering the common cases (clip path and no-clip path), matching anchor 5. | 5 / 5 |
Workflow Clarity | The loop is a clearly sequenced 6-step workflow with explicit validation checkpoints and feedback loops: doctor PASS/WAIT/FAIL gating with a retry path, a hard reproduction gate ("change no source until every issue is reproduced and its capture matches the clip frame"), and a final per-issue verification sheet ("Clip | Production before | Figma (if measured) | Local after") with recovery guidance when reproduction fails. The destructive operations (deleting production copies, changing source) are all guarded, so no workflow cap applies; this matches anchor 5's 'explicit validation steps; feedback loops for error recovery'. | 5 / 5 |
Progressive Disclosure | The written structure is textbook: the main loop and tools live in SKILL.md while detail topics are split into a clearly signaled, one-level-deep reference list ("References (open only when the clip needs them)" with a one-line description per file). However, scored against the actual bundle as the rubric directs: the shipped bundle contains only scripts/ (clip.mjs, copy-design.mjs, figma-login.mjs, export-figma-cookies.mjs), while the referenced reference/*.md files and every harness/* file the loop depends on (doctor.mjs, dlib.mjs, sheet.mjs, figlib.mjs, lock.mjs, shot.mjs, osmouse) are absent, so navigation from the body hits dead ends. That is more than anchor 5's 'easy navigation' but the signaling and split are well above anchor 3's 'references present but not clearly signaled'. | 4 / 5 |
Total | 19 / 20 Passed |