Content
47%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The skill has a well-structured four-phase workflow with clear sequencing and escalation paths, which is its strongest aspect. However, it is significantly over-verbose, repeating the core message ('don't guess, find root cause') in at least five different sections (Iron Law, Red Flags, Rationalizations, Partner Signals, Common Rationalizations). Most guidance is abstract process description rather than concrete, executable techniques, with the multi-component diagnostic bash example being a notable exception.
Suggestions
Cut the Red Flags, Common Rationalizations, and Partner Signals sections — they all repeat the same message. A single 3-line 'Anti-patterns' note would suffice.
Remove motivational/philosophical content ('Random fixes waste time', 'Symptom fixes are failure', Real-World Impact statistics) — Claude doesn't need persuading, it needs instructions.
Add concrete executable examples for Phase 2 (e.g., a git diff command sequence) and Phase 3 (e.g., a minimal test change pattern), not just the Phase 1 bash example.
Move the 'When to Use' and 'When Process Reveals No Root Cause' sections to a supporting file to keep the main skill lean and focused on the four phases.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Extremely verbose for what it teaches. Much of the content is motivational/philosophical ('Random fixes waste time', 'Symptom fixes are failure'), repeats the same point many times (red flags, rationalizations, common excuses all say 'stop guessing'), and includes things Claude already knows (scientific method, 'read error messages carefully'). The rationalizations table, red flags section, and 'your human partner's signals' section are largely redundant with each other. Could be cut by 60%+ without losing actionable content. | 1 / 3 |
Actionability | The multi-component diagnostic example with bash commands is concrete and useful. However, most of the skill is abstract process description ('Form Single Hypothesis', 'State clearly', 'Be specific') rather than executable guidance. The phases describe what to do conceptually but lack concrete code examples for most steps. The bash diagnostic example is the one bright spot. | 2 / 3 |
Workflow Clarity | The four-phase workflow is clearly sequenced with explicit gates between phases ('MUST complete each phase before proceeding'). Phase 4 includes a concrete escalation path (3+ fixes → question architecture), feedback loops (hypothesis fails → new hypothesis → return to Phase 1), and validation checkpoints (verify before continuing, create failing test). The quick reference table provides a clear summary. | 3 / 3 |
Progressive Disclosure | References to supporting files (root-cause-tracing.md, defense-in-depth.md, condition-based-waiting.md) and related skills are well-signaled at the bottom. However, the main SKILL.md itself is monolithic — the rationalizations table, red flags, and partner signals sections could be in a separate reference file. No bundle files were provided to verify referenced paths exist. | 2 / 3 |
Total | 8 / 12 Passed |