Content
100%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.
An exemplary instruction-only skill body: a tight, four-phase procedure with concrete uv commands, exact file paths, explicit validation gates, and clean delegation of standards to REVIEW.md and area guides. No dimension shows meaningful weakness.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and imperative throughout ('Do not infer behavior from a name, a docstring, or a comment', 'Do not run `just check` or the full test suite. CI runs them.'), and every line adds non-obvious project knowledge (e.g. 'The unit tests run offline with mocks', why `tests/live/` is off-limits). Nothing is padded or explains concepts Claude already knows, matching the anchor where every token earns its place. | 5 / 5 |
Actionability | Concrete executable commands with exact invocation patterns are given (`uv run python -c "..."`, `uv run pytest <file>::<test> -q`, `uv run mp help <Name>`) alongside exact file paths for each check (`src/mixpanel_headless/__init__.py`, a `CHANGELOG.md` entry under `## Unreleased`, `docs/guide/`). This is copy-paste-ready guidance fully covering the skill's common cases, and per the rubric notes the absence of code in an instruction-only skill is not penalized. | 5 / 5 |
Workflow Clarity | Four clearly sequenced phases (collect context → check each finding → check public API changes → post) with an explicit validation gate and feedback loop: 'If you cannot support the finding with a line or a check, do not post it.' Posting rules ('For a bug, state the input that fails and the wrong result', 'If the PR is clean, say so') act as a checklist, matching the top anchor; checkpoints are explicit, not implicit as at level 4. | 5 / 5 |
Progressive Disclosure | The body is under 50 lines with well-organized sections, and it appropriately delegates detail via clearly signaled, one-level-deep references — priorities, evidence standards, and severity scale live in `REVIEW.md`, and per-area guidance in `.github/instructions/` files — rather than inlining them. No bundle files exist to verify, and navigation is easy, matching the top anchor. | 5 / 5 |
Total | 20 / 20 Passed |