Content
50%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill provides a reasonable framework for reliability reviews with good structural elements (review lenses, anti-patterns, constraints, validation commands) but suffers from redundancy between sections, framework boilerplate that doesn't add review-specific value, and a gap between the conceptual guidance and truly actionable instructions. The procedure would benefit from concrete examples of what 'inspect resilience controls' looks like in practice and explicit validation checkpoints between steps.
Suggestions
Remove the duplicated 'When to use' section since the 'Use' section already covers this, and consolidate the Subagent Routing section into just a reference link since it duplicates the referenced routing file.
Add concrete, actionable examples to the Procedure — e.g., specific code patterns to grep for (missing timeout configs, retry without backoff), specific file paths to check, or a sample finding with severity/evidence/mitigation format.
Add an explicit validation checkpoint between steps 3-4 (after mapping dependencies, verify coverage before inspecting controls) and a feedback loop for expanding scope when blast radius evidence warrants it.
Provide a concrete example of a structured review output (schema_version: 1) showing the expected format for severity-ranked findings, so Claude knows exactly what to produce.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The skill has significant redundancy — 'When to use' repeats the 'Use' section, 'Full Context' and 'Subagent Routing' sections contain routing boilerplate that Claude doesn't need explained, and the 'Progressive Disclosure Entry' header adds no value. Several sections (Inputs, Outputs, Review Lenses) are reasonably tight, but overall there's notable bloat from duplicated intent and framework scaffolding. | 2 / 3 |
Actionability | The procedure provides a clear sequence of steps and the validation section has concrete bash commands, but the core review guidance remains at the level of 'inspect resilience controls' and 'map service boundaries' without concrete examples of what to look for in code, specific grep patterns, or executable analysis steps. The review lenses are a checklist of concepts rather than actionable instructions. | 2 / 3 |
Workflow Clarity | The procedure has 6 numbered steps with conditional branching (step 2 for QA reports, step 1 for loading references), and the validation section provides a fail-fast gate. However, there are no explicit validation checkpoints between review steps — e.g., no checkpoint after mapping dependencies before assigning severity, and no feedback loop for when findings need re-evaluation after expanding scope. | 2 / 3 |
Progressive Disclosure | The skill references multiple external files (resilience-patterns.md, contract.yaml, evals.yaml, routing files) with 'Read when' annotations for some, which is good structure. However, no bundle files were provided to verify these references exist, the 'Full Context' section mixes genuinely useful conditional-load references with boilerplate asset markers, and the main body still contains substantial inline content that could be offloaded (e.g., the full Subagent Routing section duplicates the referenced routing file). | 2 / 3 |
Total | 8 / 12 Passed |