Content
63%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, highly actionable body with clear mode workflows and concrete search heuristics, but it is overlong due to full sample output reports and undermines its own progressive disclosure: it cites eight bundle files that don't exist while inlining the very content it says those files hold. Trimming the sample outputs and actually splitting (or removing) the dangling references would lift both the conciseness and structure scores.
Suggestions
Replace the three full sample output reports (detection JSON, suggestion markdown, evaluation JSON, ~250 lines combined) with condensed schema sketches showing field names and one short entry each — the full reports are padding that duplicates the format contract.
Resolve the dangling references: either ship the eight cited files (reference/*.md, signatures/*.yaml, checklists/pattern-evaluation.md) and move the inlined detection rules, code-smell mappings, adaptation logic, and evaluation criteria into them, or delete the citations and keep the content inline — currently the body claims 'from signatures/...' for content that lives in SKILL.md itself.
Add an explicit validation checkpoint to the detection workflow (e.g., 'Re-read the full file before reporting a pattern with confidence <0.7 to rule out false positives'), which would also justify the existing confidence-tier gate.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is ~550 lines and mostly operational (regexes, detection rules, workflows), but the three full sample output reports (detection JSON ~65 lines, suggestion markdown ~125 lines, evaluation JSON ~60 lines) are heavily padded — condensed schema sketches would convey the same contract for a fraction of the tokens. It does not explain concepts Claude already knows, so it clears anchor 2 but fits 'mostly efficient... could be tightened' rather than anchor 4's 'minor instances'. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: named glob patterns, grep regexes (e.g., 'switch (.*type)', 'function \w+\([^)]{60,}\)'), quantified confidence bands, a stack-adaptation table, and example invocations per mode. Falls short of anchor 5 because the TypeScript examples contain placeholder bodies ('/* credit logic */') and several detection rules are deferred to signature files that are not present, leaving minor gaps. | 4 / 5 |
Workflow Clarity | Each of the three modes has a clearly sequenced numbered workflow, and the detection pipeline (Glob → Grep → Read with confidence-score bands acting as a false-positive gate) is well ordered. As a read-only skill the destructive/batch cap doesn't apply, but there are no explicit verify/re-validate checkpoints (e.g., re-reading a candidate file before reporting a pattern), which keeps it at anchor 4 rather than 5. | 4 / 5 |
Progressive Disclosure | The 'Reference files' section clearly signals one-level-deep references, but none of the eight cited files (reference/*.md, signatures/*.yaml, checklists/pattern-evaluation.md) exist in the bundle, and the body simultaneously inlines ~150 lines of that supposed reference content (detection rules, code-smell mappings, adaptation logic, evaluation criteria). This matches anchor 3 — references present, but content that should live in separate files is inline and the navigation paths are dangling. | 3 / 5 |
Total | 14 / 20 Passed |