Content
71%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 highly actionable, well-sequenced audit workflow with real bash probes, output schemas, an adversarial validation phase, and useful anti-pattern guardrails. Its weaknesses are monolithic inline templates with no bundle references, a few buggy probes, and no fallback path when detection fails.
Suggestions
Move the PROJECT-MAP.md, TECH-DEBT.md, and STARTER-KIT-PLAN.md templates into separate reference files (e.g. references/templates/) and link them one level deep, keeping SKILL.md as a lean overview — this would improve both progressive_disclosure and conciseness.
Fix the executable probe bugs: use '[ -f CAPABILITIES.md ] && head -50 CAPABILITIES.md' instead of the -d directory test, and group find predicates as 'find . \( -name "*.yaml" -o -name "*.md" \)' so the -o alternation works as intended.
Add fallback guidance for unrecognized stacks (what to do when no detection layer matches) and include the actual prompt to spawn the Phase 12 'context-validator', strengthening workflow_clarity's missing checkpoints.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with executable probes and templates and does not explain concepts Claude already knows, but the full inline templates for PROJECT-MAP.md (~80 lines), TECH-DEBT.md, and STARTER-KIT-PLAN.md, plus the cost-guardrails section, add tokens that could be trimmed or offloaded. Efficient overall with minor instances of verbosity — anchor 4, not 5. | 4 / 5 |
Actionability | Mostly copy-paste-ready bash probes (stack/topology detection, endpoint greps, convention probes) and complete output-file templates. Minor executable gaps keep it below anchor 5: '[ -d CAPABILITIES.md ] && cat CAPABILITIES.md' uses a directory test on a file, and 'find ... -name "*.yaml" -o -name "*.md"' lacks predicate grouping, so some probes misfire as written. | 4 / 5 |
Workflow Clarity | Twelve clearly sequenced phases with an explicit adversarial validation checkpoint (Phase 12, with 'blocking' → STOP feedback) and an anti-patterns checklist, satisfying validation for the batch doc-generation operations. Minor gaps remain: no fallback when stack detection matches nothing, no recovery for failed quality-baseline commands, and the context-validator prompt is never specified — anchor 4 rather than 5. | 4 / 5 |
Progressive Disclosure | Sections are well organized (Inputs, phased Workflow, Cost guardrails, Anti-patterns, Referencias), but the skill is a monolithic ~435-line file: three full output-document templates are inlined that could live in references/ files, and the 'Referencias' section points at project-destination paths (e.g. 'scripts/generate_project_map.py', '.claude/skills/pm/SKILL.md') rather than verifiable bundle files. This matches 'some structure but content that should be separate is inline'. | 3 / 5 |
Total | 15 / 20 Passed |