Content
75%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-structured skill body with excellent dispatch routing, executable commands, and deliberate one-level-deep reference deferral. The main costs are token efficiency — repeated rationales, UI minutiae, and inline version numbers — and a body that carries detail which belongs in its own references.
Suggestions
State the --quiet rationale once (in the flags table) and reference it from the Export shortcuts section and Quick usage comments instead of repeating the full explanation three times.
Move the version-annotated detail sections (subagent attribution Phase-B semantics, cross-cutting sections) into a reference file and keep a 3-5 line summary with a 'Read when' pointer in SKILL.md.
Condense the HTML drawer/interactive-detail description to the one or two facts that affect how the skill is invoked, deferring UI behavior to the HTML output itself or a reference.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and skill-specific with no generic-concept explanations, but includes unnecessary padding: the --quiet rationale is explained three times (Export shortcuts prose, the flags table, and a Quick usage comment), the HTML drawer UI description (~15 lines of toggle/focus/Esc detail) exceeds what is needed to run the skill, and inline version numbers (v1.6.0, v1.7.0, v1.36.0, v1.78.0, v1.87.0) are time-sensitive information not placed in a deprecated/old-patterns section. It is not a 2 because essentially all content is skill-specific operational fact rather than concepts Claude already knows. | 3 / 5 |
Actionability | Fully executable throughout: copy-paste-ready `uv run` commands with concrete paths and env vars, a literal-equality dispatch table for $ARGUMENTS[0], ordered first-match-wins export rules backed by a 12-row argument-string-to-command example table, and complete flag tables with defaults. This matches the anchor for copy-paste-ready guidance covering the common cases. | 5 / 5 |
Workflow Clarity | Multi-step flows are clearly sequenced with explicit gates: the dispatch gate, ordered export-scope checks, the 3-step prepare/edit/render flows for Tasks and Insights companions (render validates the prose; facts are recomputed, never trusted), scope gates ("Never auto-generate for --project-cost"), and count thresholds (2–40 request units). The destructive --prune-exports operation is dry-run by default with --yes required. It falls short of 5 because error-recovery paths (missing session JSONL, script failure, malformed exports) are implicit rather than spelled out. | 4 / 5 |
Progressive Disclosure | All eight reference files plus the two referenced scripts exist, are one level deep, and are clearly signaled with per-file "Read when..." conditions and explicit deferral rationale ("deferred so single-session reports don't pay for context they rarely use"). It is not a 5 because the body itself inlines substantial detail that could live in references — the ~25-line subagent attribution semantics section, the exhaustive flag tables, and the HTML drawer behavior description — making SKILL.md more than the lean overview the top anchor describes. | 4 / 5 |
Total | 16 / 20 Passed |