Content
56%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.
A well-structured, genuinely actionable process guide with a clear RED-GREEN-REFACTOR workflow, explicit validation checkpoints, and concrete templates and checklists. Its two real weaknesses are verbosity — extensive duplication of description guidance, the TDD cycle, and the Iron Law pushes it far past its own token targets — and broken progressive disclosure: all five referenced bundle files are missing, including the one containing the core testing methodology.
Suggestions
Consolidate duplicated material: merge the description-writing guidance and good/bad YAML examples from 'SKILL.md Structure' and 'Claude Search Optimization' into one section, fold the TDD Mapping table into the RED-GREEN-REFACTOR section, and state the Iron Law once — this alone would cut roughly a third of the body's tokens.
Ship the referenced bundle files (testing-skills-with-subagents.md, anthropic-best-practices.md, persuasion-principles.md, graphviz-conventions.dot, render-graphs.js) or inline/remove their references — currently every external link, including the core testing methodology, is a dead end.
Apply the skill's own rules to itself: replace '@testing-skills-with-subagents.md' and '@graphviz-conventions.dot' with plain named references (per 'Why no @ links'), and move the 100+ line CSO and testing-type sections into separate reference files as its 'Separate files for heavy reference' rule requires.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | At 3,308 words / 666 lines the body is over six times its own stated target ('Other skills: <500 words'), with several duplicated sections: description-writing guidance and good/bad YAML examples appear twice (SKILL.md Structure vs. Claude Search Optimization), the RED-GREEN-REFACTOR cycle is covered by both the TDD Mapping table and its own dedicated section, the Iron Law appears in both 'The Iron Law' and 'The Bottom Line', and the rationalization table is given twice (filled example plus template). This matches the anchor for noticeably verbose with several unnecessary or padded sections; it avoids a 1 because the material is domain-specific expert content rather than explanations of concepts Claude already knows. | 2 / 5 |
Actionability | The body provides mostly concrete, executable guidance: a filled-in YAML frontmatter template, explicit good/bad description examples, runnable commands ('wc -w skills/path/SKILL.md', './render-graphs.js ../some-skill --combine'), a Bad/Good counter-example pair, and a phase-organized checklist with a TodoWrite instruction. It is not a 5 because key operational detail — 'How to write pressure scenarios', 'Pressure types (time, sunk cost, authority, exhaustion)' — is deferred to testing-skills-with-subagents.md, which is not present in the bundle, so the RED phase cannot be fully executed from what is provided. | 4 / 5 |
Workflow Clarity | The RED-GREEN-REFACTOR sequence is clearly laid out with explicit checkpoints ('Run pressure scenario with subagent WITHOUT the skill. Document exact behavior', 'Run same scenarios WITH skill. Agent should now comply'), a feedback loop ('Agent found new rationalization? Add explicit counter. Re-test until bulletproof'), a phase-organized checklist, and a hard gate ('STOP: Before Moving to Next Skill'). It falls short of the 5 anchor because the actual scenario-construction methodology behind the validation steps lives in the missing @testing-skills-with-subagents.md file, leaving a gap in how validation is performed. | 4 / 5 |
Progressive Disclosure | The body has clear ## section structure, but all five referenced supporting files — anthropic-best-practices.md, graphviz-conventions.dot, render-graphs.js, persuasion-principles.md, and testing-skills-with-subagents.md — are absent from the bundle (no references/, scripts/, or assets/ directories exist), so every external reference is a dead end, including the one holding the 'complete testing methodology'. Additionally, 600+ lines of CSO guidance and testing-approach content are inlined where the skill's own rule ('Separate files for: Heavy reference (100+ lines)') says they belong in separate files, and it uses @-references ('@graphviz-conventions.dot', '@testing-skills-with-subagents.md') that its own guidance forbids. This sits between the 'references are buried/inlined content' anchor (2) and the 'good structure, some organization gaps' anchor (4): structure is genuinely good but navigation is broken, so 3 is the best fit. | 3 / 5 |
Total | 13 / 20 Passed |