Content
53%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 core reference material is well structured — executable commands, real parameters, and a genuine one-level-deep reference split — but the body is heavily padded with generic template sections (When to Use, Key Features, Implementation Details, Output Contract, Validation and Safety Rules, Failure Handling) that restate platitudes instead of CTD-specific guidance. The duplicated and vague workflow sections lack a concrete validation feedback loop.
Suggestions
Delete the boilerplate sections (Key Features, When to Use, Implementation Details, Required Inputs, Output Contract, Validation and Safety Rules, Failure Handling) or replace them with CTD-specific content; keep only Capabilities, Examples, References, and one concrete workflow.
Merge the "Example run plan" and "Recommended Workflow" into a single sequence with an explicit validation checkpoint tied to a real command (e.g., run a one-term test query and check the result before running the full batch), including a fix-and-retry step for API or validation errors.
Fix the concrete errors: remove the non-portable `cd "20260316/..."` path in Example Usage, and replace the fabricated `--help` "Expected output format" block with the script's actual usage output.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly half the body is generic template padding that adds nothing Claude doesn't know: "Use this skill when the request matches its documented task boundary", a "Key Features" section that restates the frontmatter description verbatim, abstract "Implementation Details" bullets ("Execution model: validate the request, choose the packaged workflow"), and a fabricated "Expected output format" block for `--help` that does not describe real output. It is not a 1 because the Capabilities/Examples/References sections carry genuinely useful, specific content. | 2 / 5 |
Actionability | The Capabilities section provides fully executable commands (e.g., `python scripts/query_ctd.py --inputType chem --inputTerms "Mercury" --report genes_curated --format json`), concrete parameter enumerations, and copy-pasteable direct-link URL patterns. Minor gaps keep it from 5: the Example Usage begins with a non-portable `cd "20260316/scientific-skills/Evidence Insight/ctd-api"` path, and the "Expected output format" shown for `--help` does not match what the script would actually print. | 4 / 5 |
Workflow Clarity | Two overlapping four-step lists ("Example run plan" and "Recommended Workflow") are present but generic — "Run a final validation pass for completeness, consistency, and safety" names no concrete checkpoint, and there is no validate-then-retry feedback loop despite the script driving batch API queries. The sequence is stated but checkpoints are vague, matching anchor 3; the batch-operation guidance also caps this dimension at 3. | 3 / 5 |
Progressive Disclosure | The detailed inputType/report matrix is correctly split into `references/ctd_api_docs.md` (which exists), clearly signaled as "See references/ctd_api_docs.md for a complete list of valid inputType and report combinations" and referenced again in Parameters — one level deep, no nesting. It is not a 5 because the SKILL.md overview itself is cluttered by ~10 boilerplate sections that obscure the navigation structure. | 4 / 5 |
Total | 13 / 20 Passed |