Content
57%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 dense with genuinely non-obvious domain knowledge and stays concise, but it is more of a fact sheet than an operational guide: the only code snippet is non-executable and internally inconsistent with the API listing, and no ordered workflow with validation checkpoints is given for what are budget-consuming batch optimizer runs. Referenced bundle files are also absent from the skill directory.
Suggestions
Make the Core Pattern snippet executable and consistent: show construction of a real reflection client, request, and evaluator (or point to a specific examples/ file), and reconcile the `ax.NewGEPA` vs `axllm.AxGEPA` prefix mismatch.
Add a short numbered workflow (e.g., 1. seed with BootstrapFewShot if demos needed, 2. run Optimize, 3. validate by comparing metadata.bestScore against baselineScore, 4. inspect artifact) with the comparison step as an explicit checkpoint.
Ship the referenced files (API.md, axir-api.json, axir-capabilities.json, examples/) in the bundle or remove/deweight the Package Facts pointers so no navigation path dangles.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean bullets and code — "Package: `github.com/ax-llm/ax/packages/go`", "Real network support: yes" — and nearly all of it is non-obvious domain knowledge (baseline score semantics, metric-call accounting, demo provenance) that Claude cannot already know. It stops short of anchor 5 because the opening paragraph and "When To Use" partially restate the frontmatter description, and "Relevant API Surface" is a bare name-list whose relevance is unexplained. | 4 / 5 |
Actionability | The "Core Pattern" snippet (`engine := ax.NewGEPA(reflectionClient, nil)` / `result := engine.Optimize(request, evaluator)`) is structurally concrete but not executable — the variables are undefined placeholders — and it conflicts with the "Relevant API Surface" section, which uses the `axllm.` prefix (`axllm.AxGEPA`) while the snippet uses `ax.`. This matches anchor 3 ('some concrete guidance but incomplete ... missing key details') rather than anchor 4's mostly-executable guidance; the guardrails are behaviorally concrete but give no runnable entry point. | 3 / 5 |
Workflow Clarity | Sequencing exists only implicitly: "Use BootstrapFewShot before GEPA when demonstrations should seed optimization" and "Compare metadata.bestScore with baselineScore" hint at order and verification, but no numbered workflow (bootstrap → optimize → inspect artifact → compare scores) is laid out. Optimizer runs are batch operations consuming capped metric budgets, so per the rubric's cap, missing explicit validation checkpoints holds this at anchor 3 ('sequence present but checkpoints missing or implicit') rather than 4. | 3 / 5 |
Progressive Disclosure | Sections are well-organized for a short skill, and references are clearly signaled in a dedicated "Package Facts" section, but the referenced files — `API.md`, `axir-api.json`, `axir-capabilities.json`, `examples/` — are not present in the bundle (no references/, scripts/, or assets/ directories exist), so the navigation paths dangle. This lands at anchor 3 ('structure present, references present but not resolvable/well-placed') rather than 4, since a reader following the pointers finds nothing. | 3 / 5 |
Total | 13 / 20 Passed |