Content
77%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 review workflow with explicit validation checkpoints and honest error-handling fallbacks. Its costs are token efficiency and organization: the impacted_tests contract is stated twice, and substantial response-field documentation is inlined in SKILL.md rather than split into a reference file.
Suggestions
State the impacted_tests response shape once and cross-reference it (e.g., 'get_change_risk returns the same impacted_tests block described above') instead of restating status/run_all/reasons/tests_to_run in two places.
Move the field-by-field directive documentation (may_break, missing_cochanges, the tests_* fields and their statuses) into a references/ file, keeping SKILL.md to the workflow, the primary calls, and one-line field summaries.
Tighten the parenthetical-heavy passages in the get_risk section into short field:meaning bullets so each token earns its place.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The dense tool-specific semantics (directive fields, test-selection rules) are earned knowledge Claude does not have, but the impacted_tests response shape is explained twice — once in the get_risk section and again via 'get_change_risk carries the same answer as impacted_tests' — and long parenthetical chains (the tests_to_run_why / tests_run_all / tests_to_run_basis passage) could be tightened. Mostly efficient with some unnecessary repetition. | 3 / 5 |
Actionability | Fully executable throughout: get_change_risk(revspec="main..HEAD"), get_risk(targets=..., changed_files=...), get_why(query="<file>"), get_health(targets=..., include=["biomarkers"]), gh pr diff <number>, git diff --name-only main...HEAD, and repowise impacted-tests <revspec> --format list | xargs pytest. Concrete copy-paste-ready calls cover the common cases. | 5 / 5 |
Workflow Clarity | Clear, explicitly ordered sequence ('Score the whole change first' → 'Then drill into the per-file directive' → 'Then go deeper where it matters' → 'Getting the diff' → 'Write the review') with explicit validation checkpoints: a warning field or 'unknown' status means an all-zero score is not a clean bill of health, tests_run_all=true means no smaller set can be vouched for, tests_status failure means every test must run, and an error-handling section gives fallback to the raw diff. The skill is read-only, so the destructive-operation cap does not apply. | 5 / 5 |
Progressive Disclosure | No bundle files exist and the skill is a single ~113-line file with clear section headers, but the field-by-field response documentation (may_break, missing_cochanges, the tests_* fields) and the duplicated impacted_tests block are inline API-reference material that would sit better in a separate reference file. Some structure, but content that should be separate is inline. | 3 / 5 |
Total | 16 / 20 Passed |