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 well-organized, token-efficient skill body with genuinely executable commands and concrete thresholds, undermined by bundle-integrity failures: the 'Loading order for every review' mandates two files ('rules/universal.md', 'languages/*.md') that are not shipped, and the regression-fixture validation step depends on a missing 'expected_outputs/' directory. The design of the disclosure structure is strong, but as delivered the primary navigation paths are broken.
Suggestions
Ship the referenced 'rules/universal.md' and 13 'languages/*.md' files (or remove the loading-order section and inline the universal rules) — the mandated workflow currently directs the agent to files that do not exist in the bundle.
Commit the 'expected_outputs/*.json' files the regression-fixture section depends on, or drop the section — the documented diff command fails as shipped since only the 'assets/' fixtures exist.
Collapse the duplication between the directory tree (which enumerates all 13 language files) and the extension dispatch table — one listing of the same mapping is enough.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and functional — command examples, threshold and verdict tables, and dispatch rules with no padding or explanations of concepts Claude already knows. Minor trimmable redundancy: the 'How This Skill Is Organized' tree lists all 13 'languages/*.md' files individually, then the extension dispatch table repeats the same mapping, and 'That is always exactly 2 additional files' restates the loading order. | 4 / 5 |
Actionability | Commands are concrete and verified executable against the bundled scripts (e.g. 'python scripts/pr_analyzer.py . --base main --head feature-branch', '--pr-analysis pr_results.json', all matching real argparse flags), and 'Adding a New Language' cites real symbols ('LANGUAGE_EXTENSIONS', 'check_<name>_specific_smells(...)'). The gap is more than cosmetic but confined: the mandated loading of 'rules/universal.md' and a 'languages/*.md' file cannot be executed because neither directory ships with the bundle, leaving the promised 'Language-specific detections' and 'Language-specific checks' undefined. | 4 / 5 |
Workflow Clarity | The 'Loading order for every review' is an explicit 1-2-3 sequence with a dispatch table, the verdict table defines decision criteria, and the regression-fixture section provides a verification loop ('Drift from the committed JSON signals a behaviour change'). Not 5 because that verification loop references a nonexistent 'expected_outputs/' directory, so the checkpoint cannot actually run as written. | 4 / 5 |
Progressive Disclosure | The hub-and-spoke design is structurally excellent — SKILL.md as dispatch table pointing one level deep to 'rules/universal.md' and 'languages/*.md', clearly signaled by extension — but scoring against the actual bundle, every one of those 15 referenced rule files is absent, so the core progressive-disclosure structure delivers nothing. Not 4 because 'references mostly clear' with only 'minor organization gaps' understates broken references on the primary loading path; not 2 because the inline content that does exist (dispatch table, thresholds, tool docs) is well organized. | 3 / 5 |
Total | 15 / 20 Passed |