Content
67%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 delivers excellent, actionable project-specific constraints (async auto mode, module-level imports, in-memory transport, result.data accessor, inline-snapshot workflow) but dilutes them with substantial generic pytest guidance Claude already knows. Structure and actionability are solid; the main improvement is cutting or relocating the generic content.
Suggestions
Cut or move generic pytest sections (Test Naming, Error Testing, tmp_path, basic mocking, function-scoped fixtures) to a references file, keeping SKILL.md focused on the project-specific rules Claude cannot infer.
Replace placeholder-dependent snippets (User, calculate, my_function, MyModel) with runnable examples using real project imports like the FastMCP in-memory transport example.
Add a short explicit workflow (write test -> uv run pytest -n 2 -x -> fix or --inline-snapshot=fix -> re-run) to turn the implied feedback loop into a checkpointed sequence.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The project-specific rules (asyncio_mode auto, module-level imports, in-memory transport, result.data vs result[0].text, inline-snapshot) are high-value, but roughly half the body restates generic pytest knowledge Claude already has (descriptive test names, pytest.raises with match, tmp_path, mock-at-the-boundary, function-scoped fixtures, basic parameterization). It fits 'mostly efficient but includes some unnecessary explanation' better than the padded anchor at 2, since the writing itself is terse. | 3 / 5 |
Actionability | Good/bad code pairs, copy-paste run commands ("uv run pytest -n 2 --inline-snapshot=create"), and a complete in-memory transport example make the guidance mostly executable. It falls short of 5 because several snippets use undefined placeholders (User, calculate, my_function, MyModel), making them illustrative patterns rather than fully executable code. | 4 / 5 |
Workflow Clarity | Sections are clearly organized with a pre-submit checklist and a snapshot create-then-fix command flow, providing most checkpoints; no destructive or batch operations apply, so no cap is triggered. It is not 5 because the write-test -> run -> fix feedback loop is implied by the run commands but never laid out as a sequence. | 4 / 5 |
Progressive Disclosure | No bundle files exist, and the single ~230-line file is well-sectioned with one clearly signaled external reference. The generic pytest sections (Test Naming, Error Testing, Fixtures, Mocking) could be split into a reference file to keep SKILL.md focused on project constraints, matching the minor-organization-gaps anchor at 4 rather than the ideal split at 5. | 4 / 5 |
Total | 15 / 20 Passed |