Content
66%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 well-architected index: crisp principles, a decision table, gated phases with validation and a bounded feedback loop, and a Definition of Done. Its fatal flaw in this bundle is that the progressive-disclosure chain is broken — almost every referenced rule/template file is absent, so the thin-index design points at nothing. Secondary issues are duplicated file-link tables and verbose 7.5/7.6 cells.
Suggestions
Ship the referenced files: the entire `rules/` directory (17+ files), `templates/`, and `lens.md` are missing from the bundle, so nearly all links in the decision flow, workflow table, and required-reading table dead-end.
Merge the "Required Reading by Phase" table into the Workflow table — both repeat the same phase → rule-file mapping, doubling the link maintenance surface.
Compress the Phase 7.5 and 7.6 cells to one-line pointers to `rules/record-and-iterate.md`, since the cells currently restate that file's full procedure.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is high-signal and opinionated ("Animate `transform`, `opacity`, and `filter` only", "150–500 ms band for UI") with no concept explanations Claude already knows, but it could be tightened: the "Required Reading by Phase" table repeats the same rule-file links already given in the Workflow table, and the Phase 7.5/7.6 cells spell out full procedures ("invoke `Skill(\"screencast\")` twice — once with `\"reducedMotion\": false\`, once with `true`...") that duplicate [`rules/record-and-iterate.md`] they also link. Mostly efficient with some duplication — anchor 3, not 4's "minor instances". | 3 / 5 |
Actionability | Concrete, executable direction throughout: a first-match-wins decision table mapping signals to exact tools ("`animation-timeline: scroll()` / `view()`", "`interpolate-size: allow-keywords`"), numeric duration bands, a runnable command ("`node ${CLAUDE_SKILL_DIR}/scripts/is-motion-diff.mjs`" — the script exists), and a Definition of Done checklist. Not 5 because copy-paste code lives in [`references/recipes.md`] and [`templates/`] and most deep guidance sits behind rule files that are not present in the bundle, leaving minor gaps for the common cases. | 4 / 5 |
Workflow Clarity | The phased workflow (A, B, 0–7.6) is clearly sequenced with an explicit gate per phase ("Animated property is `transform`, `opacity`, or `filter`. If not, justify with a layout-thrash measurement"), measurement validation (Phase 7: "no purple Layout / green Paint bars"), and a real feedback loop with a bounded retry and escalation path ("apply the fix and return to Phase 7.5. Cap the loop at 3 iterations — escalate via `Skill(\"confidence\", analysis)` on the 4th"), plus a closing checklist. Matches the anchor-5 example pattern of validate → fix → re-validate. | 5 / 5 |
Progressive Disclosure | The index design is exemplary on paper — "This `SKILL.md` is a thin index... Detailed rules live in `rules/*.md` and load on demand", one level deep, clearly labeled — but scored against the actual bundle, nearly every referenced path is dead: `rules/` (17+ files), `templates/`, and `lens.md` do not exist; only `references/recipes.md` and `scripts/is-motion-diff.mjs` are present, and recipes.md itself links to the missing `../rules/state-choreography.md`. Navigation dead-ends on first use, which lands below the midpoint; not 1 because the structure is well organized, not a monolithic wall. | 2 / 5 |
Total | 14 / 20 Passed |