Content
71%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 tightly written, well-sequenced instruction skill that earns its tokens and encodes real craft standards, with clear completion criteria and a verification pass before handoff. Its one significant defect is that both the Hard Rules and Phase 4 direct the model to load and copy PICKER.md verbatim, and that file is absent from the bundle, leaving the skill's central artifact unspecified.
Suggestions
Ship PICKER.md alongside SKILL.md (e.g., in references/) containing the picker's exact markup, styles, and keyboard wiring, or inline that spec in Phase 4 if the bundle is meant to stay single-file.
Add an explicit error-recovery loop to Phase 5 (e.g., "if a variant fails to render or throws console errors, fix it and re-flip before presenting") to earn the top workflow-clarity anchor.
Provide a minimal starter for the no-project branch (a skeleton of the self-contained HTML harness) so the standalone path is as executable as the in-project path.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence throughout — motion standards are stated as facts ("sub-300ms UI motion", "transform/opacity only", "never `ease-in`") rather than explained, and there is no padding of known concepts. Every section drives behavior: hard rules are terse imperatives, phases carry exact counts and completion criteria, and the invocation grammar is given as syntax. Re-reading anchor 4, there are no real instances of over-explanation to trim, so the lean anchor fits best. | 5 / 5 |
Actionability | Most guidance is concrete and executable ("an isolated route or page (`/prototypes/<slug>`, or the framework's equivalent)", the invocation-variant table, the defined presentation-table format), but the most load-bearing artifact is delegated to a missing file: "Its exact markup, styles, and behavior are specified in [PICKER.md](PICKER.md) — copy them verbatim" and "load it now and build exactly that" both point at a PICKER.md that does not exist anywhere in the bundle (no references/ directory). That is a missing key detail, matching anchor 3; it is not 4 because the gap breaks a Hard Rule and a phase, and not 2 because the remaining guidance is specific and directive rather than high-level hints. | 3 / 5 |
Workflow Clarity | Six clearly sequenced phases with named checkpoints — "**Completion criterion:** every variant has a name and a stated axis", and Phase 5's "Run the harness. Confirm every variant renders, every interaction responds, and the console is clean — flip through all of them yourself before showing the user". It falls short of anchor 5 only because there is no explicit error-recovery loop (what to do when a variant fails verification); the only destructive step (deleting the prototype surface in Phase 6) is guarded by user selection and Hard Rule 1, so no destructive-operation cap applies. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned and makes the right disclosure move — pushing the bulky picker spec out of SKILL.md to a one-level-deep, prominently signaled reference ("specified in [PICKER.md](PICKER.md)") rather than inlining it. But the bundle listing shows no references/, scripts/, or assets/ directories and no PICKER.md anywhere, so the disclosure chain is broken at its only link, which keeps it at anchor 3 rather than 4–5; it is above 2 because nothing is wrongly inlined and the reference is clearly signaled, not buried. | 3 / 5 |
Total | 15 / 20 Passed |