Content
57%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.
The body is highly actionable — exact commands, scoring weights, scripted analysis steps, and explicit user-confirmation gates — and it correctly points to a real bundle script. Its main weaknesses are the absence of any output-validation checkpoint in a long-running batch workflow, incoherent step numbering (Phase 3 / missing Step 4), and heavy inlining of reference-grade material (CSV schemas, interpretation guide) that should live in separate one-level-deep files.
Suggestions
Add a validation step after running the tool, e.g., 'Check the output CSV exists and has the expected row count; if the run failed or was rate-limited, report the error and retry with a smaller --limit', to lift workflow_clarity above the batch-operation cap.
Move the two CSV column tables and the 'Output Interpretation Reference' into a references/OUTPUT_SCHEMA.md and link to it, keeping SKILL.md as a lean overview.
Fix the step numbering: renumber the post-run steps consistently (Steps 1-8 under one scheme) and remove the orphaned 'Phase 3' header so the sequence reads coherently.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly operational instruction rather than concept explanation, but it can be tightened: the two full CSV column tables ('repo_signals_users.csv — One row per person...', 'repo_signals_interactions.csv — One row per user x repo combination') and the 'Output Interpretation Reference' section are reference material inlined in SKILL.md, and the scoring-weights list duplicates what the script already implements. Not 2 because there is no padding that explains concepts Claude already knows; not 4 because several sections are reference-grade content that adds tokens without being needed at execution time. | 3 / 5 |
Actionability | Concrete, executable guidance throughout: the exact command 'python3 ${CLAUDE_SKILL_DIR}/scripts/gh_repo_signals.py --repos ... --limit ... --output ...' (verified: scripts/gh_repo_signals.py exists in the bundle), 'gh auth status', explicit scoring weights, exact user-facing questions to ask, and specific cost formulas ($0.05/$0.20 per enrichment). Not 5 because of minor gaps: the '--output .../repo_signals.csv' argument's relationship to the two produced files ('_users.csv' and '_interactions.csv') is left implicit, and 'Estimate credit cost: N users x cost per enrichment call' leaves the agent to supply the rate. | 4 / 5 |
Workflow Clarity | A clear sequence exists (verify environment -> run tool -> review output -> collect context -> analyze -> recommend -> ask go-ahead) with some checkpoints ('Do NOT proceed to analysis until you have this context', 'Wait for user confirmation before spending any credits'). However, this is a batch operation (hundreds to thousands of API calls producing CSVs) with no step validating that the run succeeded or that the output files contain expected data before analysis, which caps workflow clarity at 3 per the rubric. Numbering is also incoherent: 'Execution Steps' ends at Step 3, a 'Phase 3' header appears mid-flow, and Step 4 is missing entirely. | 3 / 5 |
Progressive Disclosure | The body has good section headers and correctly references the real bundle script (${CLAUDE_SKILL_DIR}/scripts/gh_repo_signals.py exists alongside gh_common.py, gh_contributors.py, etc.), but there is no references/ layer at all: the CSV column schemas, the output-interpretation guide, and the long conditional recommendation framework are all inlined in a ~220-line SKILL.md instead of being split into one-level-deep reference files. Not 2 because the content is well-sectioned and navigable; not 4 because content that clearly belongs in separate reference files is inline. | 3 / 5 |
Total | 13 / 20 Passed |