Content
47%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 comprehensive and well-structured evaluation framework with clear scoring rubrics and phased workflows, but it severely violates its own principles of conciseness and lean context. At 400+ lines, it far exceeds the 80-300 line guideline it recommends for skills, and embeds extensive reference material (research citations, best practices, remediation patterns) that should be in bundled files. The content is actionable at a process level but lacks executable tooling.
Suggestions
Reduce SKILL.md to 150-250 lines by moving the Lean Context Principle section, Size Guidelines Reference, Skill Quality Dimensions, Skill Development Best Practices, and Remediation Patterns into separate bundled reference files (e.g., lean-context.md, size-guidelines.md, remediation-patterns.md)
Remove or drastically trim the SkillsBench research citations and Anthropic guidance summaries—these are background justification, not actionable evaluation instructions
Cut the noise/signal tables in the Lean Context section; a brief 2-line summary with a reference to a bundled file would suffice
Add concrete executable examples: a sample script or command sequence for scanning artifacts, counting lines, or validating frontmatter rather than relying on natural language prompts
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | This skill is extremely verbose at ~400+ lines. It extensively explains concepts Claude already knows (what agentic patterns are, how folder structures work, what YAML frontmatter is). It includes lengthy sections on 'Lean Context Principle' with noise/signal tables, SkillsBench research citations, and best practices that are general knowledge rather than actionable evaluation steps. The irony is that a skill about lean context is itself full of noise. | 1 / 3 |
Actionability | The evaluation workflow provides concrete scoring tables with specific point values and criteria, which is actionable. However, there are no executable code examples—no scripts, no validation commands, no concrete file-parsing logic. The 'Running the Evaluator' section just shows natural language prompts rather than concrete implementation. The scoring criteria are specific enough to act on but lack executable tooling. | 2 / 3 |
Workflow Clarity | The 7-phase evaluation workflow (Discovery Scan → Foundation → Skills → Agents → Instructions → Consistency → Generate Report) is clearly sequenced with explicit scoring criteria at each phase. Each phase has a detailed rubric table with point allocations. The report template provides a clear output format with prioritized issues (P0/P1/P2) and remediation steps. | 3 / 3 |
Progressive Disclosure | The skill references supporting files (checklist.md, report-template.md) and mentions a related skill (project-scaffold), but no bundle files are provided. The SKILL.md itself is monolithic—embedding extensive reference material (size guidelines, quality dimensions, remediation patterns, best practices, lean context principle) that should be in bundled files. The skill violates its own advice about keeping SKILL.md to 1-2K tokens. | 2 / 3 |
Total | 8 / 12 Passed |