Content
75%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 highly actionable tool reference with executable commands, concrete parameter tables, and a verified bundle structure. Its main weakness is redundancy: the same environment-variable resolution and parameter-priority material is repeated across three sections, some of which also duplicates references/api_spec.md.
Suggestions
Consolidate the three repeated env-var resolution tables (sn-image-recognize, sn-text-optimize, and 'Default Parameter Behavior') into a single resolution-order section.
Move the detailed parameter tables that duplicate references/api_spec.md out of SKILL.md, keeping only the per-tool minimal usage examples inline.
Trim the 'Mapping Between base-url and Interface Type' table (the LLM and VLM rows are identical) and the 'VLM vs LLM' table, which restates what the tool sections already say.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The env-var resolution chains ("SN_VISION_API_KEY -> SN_CHAT_API_KEY -> SN_API_KEY") are documented three separate times — in the sn-image-recognize table, the sn-text-optimize table, and again in 'Default Parameter Behavior' with added prose ("api_key resolution order (high to low): CLI --api-key > command-specific key..."). Content is dense reference tables rather than padded concept explanation, so it is above anchor 2, but the substantial duplication keeps it below anchor 4. | 3 / 5 |
Actionability | Every tool has copy-paste-ready commands ("python scripts/sn_agent_runner.py sn-image-generate --prompt \"...\"") covering minimal calls, base-url/api-key overrides, and Anthropic-compatible invocations, plus concrete success and failure JSON examples with "status" and "elapsed_seconds". Fully executable and the examples cover the common cases. | 5 / 5 |
Workflow Clarity | Clear sequence from "pip install -r requirements.txt" through per-tool invocations to documented success/failure outputs ("status": "ok" / "status": "failed") that act as verification checkpoints. Not 5 because no explicit error-recovery guidance appears in the body (error handling is deferred to references/api_spec.md); not 3 because each tool is an unambiguous single command and output verification is documented. | 4 / 5 |
Progressive Disclosure | The body signals a real, one-level-deep reference ("See `references/api_spec.md` for details") that exists and is well-organized, and all script paths cited in examples exist under scripts/. Not 5 because SKILL.md duplicates a fair amount of the reference's parameter documentation (the 'Default Parameter Behavior' section restates the per-tool tables), so the split between overview and detail is not clean. | 4 / 5 |
Total | 16 / 20 Passed |