Content
81%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 well-structured evaluation skill with genuinely actionable procedure, templates, and a concrete worked example, plus well-signaled one-level-deep cross-references. Its main cost is redundancy: the load-bearing keep-it rule is repeated across nearly every section, and the two decision templates overlap heavily.
Suggestions
State the load-bearing/keep-it decision rule once (e.g., in the Procedure's step 6) and remove the repeated restatements in the opening sections and the post-procedure paragraph.
Differentiate or merge the Decision Template and UI Refactor Template, which share the product-sentence, deletion-prize, and decision fields verbatim — the UI variant only needs to add its 'Must preserve' / 'Can refuse' lists.
Move the UI Refactor Template, UI Artifact Ladder, and worked example into a reference file (e.g., references/ui-refusals.md) to shorten SKILL.md to its spine, procedure, and core template.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The core decision rule is restated four-plus times: "keep it when the evidence says the loss is load-bearing", "still be worth keeping when the capability it provides is load-bearing", "If the loss is load-bearing, keep the feature", and "Keep the feature when the user loss is load-bearing" (plus the fidelity restatement at the end). The Decision Template and UI Refactor Template also share four fields verbatim. This is more than the minor trimmability of anchor 4 — genuine tightening is available — but there is no padding or explanation of concepts Claude already knows, so it stays above anchor 2. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is fully executable: a six-step procedure with concrete candidate types ("fast paths, old shapes, rare modes, provider exceptions, compatibility aliases, fallback parsers"), itemized deletion-prize evidence ("methods, adapters, unions, error variants, tests, docs branches"), two copy-paste-ready fill-in templates, a five-rung artifact ladder with use conditions, and a worked example naming real artifacts ("signInWithIdToken", "OIDCProvider narrowing", "two social sign-in docs branches"). Specific examples cover the common cases. | 5 / 5 |
Workflow Clarity | The Procedure is a clearly sequenced six steps with explicit checkpoints: step 4 separates purpose-changing candidates, step 5 states the smallest-loss preference rule, and step 6 is an explicit decision gate ("If the loss is load-bearing, keep the feature... Otherwise, refuse... write that refusal into the spec"). Not a destructive or batch operation, so no validation cap applies; the checkpoints are explicit rather than implicit, fitting anchor 5 over anchor 4. | 5 / 5 |
Progressive Disclosure | Clear section headers throughout, and every external pointer is one level deep with a stated purpose ("[Refactoring methods](../collapse-pass/references/refactoring-methods.md) guide caller counts, fixtures, state branches...", Rethink's clean-break reference, the docs article for narrative context). However, at ~218 lines the UI Refactor Template, UI Artifact Ladder, and worked example — roughly half the file — are inline content that could be split into reference files, and the under-50-line simple-skill exception does not apply, so it fits anchor 4's 'minor organization gaps' rather than anchor 5's 'content appropriately split'. | 4 / 5 |
Total | 17 / 20 Passed |