Content
40%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 well-sectioned and backed by a real bundle of reference, asset, and script files, but it spends most of its tokens on generic prompt-engineering knowledge Claude already possesses, presents a non-runnable fictional-library example as its Quick Start, and defines no concrete workflow with validation checkpoints. It functions as a topical overview rather than an operational skill.
Suggestions
Replace the fictional `prompt_optimizer` Quick Start with a runnable example — either plain string/f-string prompt construction or an invocation of the bundled `scripts/optimize-prompt.py` showing its actual CLI and expected output.
Cut the Best Practices, Common Pitfalls, Performance Optimization, and Next Steps sections (all generic knowledge) and fold anything non-obvious into the existing reference files, keeping SKILL.md as a lean overview with decision guidance on which reference to open.
Fix the broken pointer at line 23 (`resources/implementation-playbook.md` does not exist) to reference a real file in `references/`, and define a concrete iteration loop (draft → run `scripts/optimize-prompt.py` → compare metrics → refine) with explicit validation checkpoints.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Multiple sections restate knowledge Claude already has at a generic level — "Best Practices" ("Be Specific: Vague prompts produce inconsistent results"), "Common Pitfalls", "Performance Optimization" bullets ("Remove redundant words and phrases", "Cache common prompt prefixes"), and a "Next Steps" section — all padded with no skill-specific information. This is noticeably verbose rather than merely having some slack, matching the 2 anchor; it is not a 1 because there is no tutorial-style explanation of basic concepts. | 2 / 5 |
Actionability | The Quick Start is effectively pseudocode: it imports a fictional `prompt_optimizer` library (`PromptTemplate`, `FewShotSelector`) whose API is never explained or installable, and the Integration Patterns are f-string sketches. The rest of the body is abstract bullets ("Example selection strategies (semantic similarity, diversity sampling)") without concrete how-to, matching the 'pseudocode instead of executable code' anchor. Not a 4 because no snippet in the body can actually run or be followed verbatim. | 3 / 5 |
Workflow Clarity | The closest thing to a workflow is the Instructions list — "Clarify goals... Apply relevant best practices and validate outcomes" — which is a rough, poorly defined sequence with no concrete steps and no validation checkpoints (nothing says how to validate or what to do on failure). This matches the 'rough sequence present but many gaps; validation absent' anchor rather than the 3 anchor, whose steps are at least concretely defined. | 2 / 5 |
Progressive Disclosure | The Resources section cleanly signals one-level-deep references (five references/, two assets/, one scripts/ file — all verified to exist), and the body is organized with clear sections. Two gaps keep it from 5: the body's own pointer "open `resources/implementation-playbook.md`" references a path that does not exist (the directory is `references/`, which contains no such file), and the Core Capabilities / Key Patterns sections partially duplicate topics already covered by dedicated reference files. | 4 / 5 |
Total | 11 / 20 Passed |