Content
27%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
This skill is massively over-engineered and verbose, explaining numerous software engineering concepts Claude already knows (SOLID, Clean Architecture, DDD, common libraries, testing patterns). The theatrical 'identity' framing wastes significant tokens. While it provides some structure (triage levels, checklists, report template), the actionable content is buried in hundreds of lines of generic advice that should either be omitted or split into referenced files.
Suggestions
Reduce content by 70-80%: Remove explanations of concepts Claude already knows (SOLID, DDD, code smells, common libraries, testing patterns) and focus only on the specific reflection workflow steps and report format.
Split into multiple files: Extract the code-specific criteria, fact-checking checklist, anti-patterns catalog, and report template into separate referenced files, keeping SKILL.md as a concise overview with clear navigation.
Remove the theatrical identity/threat framing entirely - it wastes tokens and doesn't improve output quality. Replace with a single sentence about maintaining high standards.
Make the triage system actionable with concrete examples: instead of vague categories like 'simple tasks', provide specific triggers (e.g., 'if diff touches <3 files and <50 lines → Quick Path').
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Extremely verbose at ~500+ lines. Explains concepts Claude already knows extensively (what Clean Architecture is, what SOLID principles are, what code smells are, common libraries like lodash/date-fns, what AAA testing pattern is). Massive amounts of padding with generic software engineering advice that doesn't earn its token cost. The 'identity' section with threats ('you will be killed') is wasteful theatrical framing. | 1 / 3 |
Actionability | Provides checklists and a report format template which are somewhat concrete, but the code examples are trivial illustrations (date formatting) rather than executable guidance for the actual task of self-reflection. The skill is mostly abstract meta-instructions ('evaluate your output against these criteria') rather than specific, copy-paste-ready procedures. The pseudocode decision framework ('IF common utility → Use established library') is vague. | 2 / 3 |
Workflow Clarity | There is a multi-step workflow (triage → assessment → refinement → verification) with some sequencing, but the steps are bloated and lack clear validation checkpoints between them. The triage system (Quick/Standard/Deep) is defined but the thresholds are subjective. The 'decision point' at Step 2 is good but the overall flow is buried in excessive detail, making it hard to follow as an actual workflow. | 2 / 3 |
Progressive Disclosure | Monolithic wall of text with no references to external files and no bundle files provided. Content that could be split into separate reference files (code-specific criteria, fact-checking guidelines, report template, anti-patterns catalog) is all inlined, creating an enormous single document. No navigation aids or clear signposting between major sections. | 1 / 3 |
Total | 6 / 12 Passed |