Content
77%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This is a strong, well-structured review skill with excellent actionability and workflow clarity. The six-pass framework with explicit severity classifications and a fixed report template gives Claude unambiguous guidance. Minor weaknesses are slight verbosity in some passes (particularly Pass 4's extended deliberation-detection guidance) and the lack of bundle files to offload reference material, though the single-file approach is defensible given the skill's read-only nature.
Suggestions
Consider extracting the Pass 4 ambiguity trigger phrases and rewrite rules into a separate reference file (e.g., AMBIGUITY-PATTERNS.md) to keep the main skill leaner and more scannable.
Tighten the 'Do not' section — the read-only constraint is already stated in the intro and 'When to use' sections, so the repetition could be reduced to a single concise reminder.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill is thorough and mostly earns its length given the complexity of a six-pass review process, but some sections are slightly verbose — e.g., the extended explanation of mid-deliberation prose in Pass 4 with the rewrite rule and trigger phrases could be tightened. The 'Do not' section repeats constraints already stated in the intro. Overall reasonably efficient but not maximally lean. | 2 / 3 |
Actionability | Highly actionable: each pass has concrete criteria with specific examples (e.g., 'money modelled as float → violates pence-as-integer rule and ADR 038'), exact file paths to read, severity classifications, and a verbatim report template with fixed verdict options. Claude knows exactly what to do at every step. | 3 / 3 |
Workflow Clarity | The workflow is clearly sequenced: load context (5 numbered steps) → run six named passes with explicit severity rules → produce a structured report with deterministic verdict logic. Validation is built into the process (each pass is itself a validation checkpoint), and the verdict rules provide clear decision criteria including escalation paths (Blocked → revise-spec or grill-me). | 3 / 3 |
Progressive Disclosure | The skill is a single monolithic file with no bundle files. While the content is well-organized with clear section headers, the Pass 4 ambiguity section in particular contains substantial inline detail (trigger phrases, rewrite rules, examples) that could be split into a reference file. For a skill of this length (~180 lines of substantive content), some progressive disclosure into supporting files would improve scannability. | 2 / 3 |
Total | 10 / 12 Passed |