Content
77%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 disciplined, dense index with an excellent decision flow, per-phase gates, a real record→analyse→iterate feedback loop, and a Definition of Done checklist — workflow clarity is near best-in-class. The dominant defect is against the actual bundle: the carefully built progressive-disclosure structure points mostly at rule and template files that are not shipped, so the promised on-demand detail is unreachable except for references/recipes.md, and the two phase tables duplicate each other.
Suggestions
Ship the referenced bundle files: the ~15 `rules/*.md` files and `templates/cursor-spotlight.html` cited throughout the decision flow, workflow, and required-reading tables are absent from the bundle — either include them or fold their essential gates into SKILL.md so `references/recipes.md` is not the only reachable deep-dive.
Merge the "Workflow" and "Required Reading by Phase" tables into a single phase → file → gate table; both currently repeat the same phase-to-file mapping at a cost of ~20 lines.
Trim the Phase 7.5/7.6 gate cells to one-line pointers into `rules/record-and-iterate.md` — the 100+ word inline integration spec (recorder invocation arguments, analyser findings, iteration cap, escalation) duplicates what that referenced file is meant to carry.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body never explains concepts Claude already knows — it opens with the rule "Animate `transform`, `opacity`, and `filter` only" and stays at that altitude throughout, with no tutorial padding. Two things could be trimmed: the "Workflow" and "Required Reading by Phase" tables duplicate the same phase→file mapping, and the Phase 7.5/7.6 gate cells run 100+ words of inline integration detail ("invoke `Skill("screen-recorder")` twice — once with `reduced-motion: false`, once with `reduced-motion: true`...") that duplicates the referenced rules/record-and-iterate.md. That is minor over-content, matching anchor 4 rather than the every-token-earns-its-place anchor 5. | 4 / 5 |
Actionability | The in-body guidance is concrete and followable: a first-match-wins decision table ("One-shot state change on hover, focus, or class toggle → CSS transition"), specific thresholds ("Duration is in the 150–500 ms band", "the 200 ms loader floor"), exact skip-line formats ("screen-recorder — skipped (not installed)"), and a Definition of Done checklist. It stops short of anchor 5 because no executable code lives in the body and several procedure steps defer entirely to rule files that are not present in the bundle (e.g. "Look up the interaction in the catalog" — the catalog itself is not shipped), leaving minor-to-moderate gaps in what can actually be executed as written. | 4 / 5 |
Workflow Clarity | Phases are explicitly ordered ("Walk these in order. First match wins." / "For any animation task — author or review — walk these phases"), each phase has a pass/fail gate, and Phase 7.5→7.6 is a genuine feedback loop with error recovery: "If a finding contradicts the animation contract... apply the fix and return to Phase 7.5. Cap the loop at 3 iterations — escalate via `Skill("confidence", analysis)` on the 4th." Together with the "Definition of Done" checklist this matches the anchor-5 pattern of explicit validation, feedback loops, and checklists. | 5 / 5 |
Progressive Disclosure | The thin-index design is exemplary on paper — "This `SKILL.md` is a thin index. Detailed rules live in `rules/*.md` and load on demand", with clearly signaled one-level references and a per-phase file map. But scored against the actual bundle, only `references/recipes.md` (397 lines of worked recipes) ships: none of the ~15 referenced `rules/*.md` files nor `templates/cursor-spotlight.html` exist in the bundle, so the majority of references dangle and the load-on-demand structure cannot actually be followed. This lands between anchor 2 (references effectively unusable) and anchor 4 (minor organization gaps). | 3 / 5 |
Total | 16 / 20 Passed |