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.
The body is a strong, executable audit procedure: concrete grep patterns, explicit Read-verification checkpoints, a clearly sequenced five-phase workflow, and a well-specified output format with false-positive guards. Its main weaknesses are minor padding (duplicate Scope section, invocation examples) and missed opportunity to split the detection-pattern catalog into a reference file.
Suggestions
Remove the duplicate '## Scope' section at the end (it repeats the opening '**Scope**' line) and consider trimming 'Invocation Examples', which mostly restates the description's triggers.
Move the 10-pattern detection catalog or the Phase 4 cross-reference table into a references/ file, keeping SKILL.md as a shorter overview with clearly signaled pointers.
Convert the judgment-call 'Search' entries (patterns 4, 9, 10) into concrete grep patterns so every detection step is mechanically executable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Largely lean and table-driven with no explanation of concepts Claude already knows, but includes trimmable padding: a duplicate '## Scope' section (the opening '**Scope**' line plus the trailing '## Scope' section), an 'Invocation Examples' section, and a somewhat preachy 'Tool Use Is Mandatory' block. Efficient with minor instances of over-explanation — anchor 4, not 5. | 4 / 5 |
Actionability | Provides verbatim grep patterns (e.g., 'NavigationStack\s*(\(\s*\))?\s*\{', '.navigationDestination(for:', '.onOpenURL'), concrete per-pattern fixes, and fully specified output templates. Mostly executable guidance; minor gaps — some 'Search' entries (pattern 4 'NavigationStack in master-detail contexts') are judgment calls rather than literal patterns, and fixes are one-line directions rather than code examples. Anchor 4. | 4 / 5 |
Workflow Clarity | Five clearly sequenced phases (map architecture → detect anti-patterns → reason about completeness → cross-reference → score) with explicit validation checkpoints: 'For every grep match, use Read to verify the surrounding context before reporting' and 'Require evidence from the Phase 1 map — don't speculate.' The false-positive exclusion list and output limits further guard against erroneous reporting. This is a read-only audit, so no destructive/batch validation cap applies. Anchor 5. | 5 / 5 |
Progressive Disclosure | No bundle files exist, and the body is a single well-sectioned document with clear headers, one-level-deep cross-skill references only (clearly signaled: 'use `swiftui-performance-analyzer`', 'see `axiom-swiftui` skill'), and no nested references. The 10 inline detection patterns and the long cross-reference table could plausibly live in reference files for a ~265-line document, keeping this at anchor 4 rather than 5. | 4 / 5 |
Total | 17 / 20 Passed |