Content
78%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 a highly actionable, executable set of API workflows with clear sequencing and a genuinely useful one-level-deep ADMET reference. Its main weaknesses are moderate: the shipped scripts are invisible from SKILL.md (and their functionality is duplicated inline), and the description of capabilities in the body is broader than the frontmatter advertises.
Suggestions
Reference the bundled scripts in the body (e.g. in each workflow: "For batch use, see scripts/chembl_target.py and scripts/ro5_screen.py") instead of duplicating their functionality inline — this would shrink SKILL.md and make all three bundle files discoverable.
Add a short error-recovery note for API failures (HTTP errors, empty JSON, rate-limit 429s) to close the workflow validation gap; the scripts already model this behavior.
Trim the persona opener and the "world's largest open bioactivity database" aside — they add tokens without adding guidance Claude does not already have.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Nearly every token is executable code with concrete example IDs ("CHEMBL203", "CHEMBL25 (aspirin)"), with only minor trimmable padding: the "You are an expert pharmaceutical scientist" persona opener, the "world's largest open bioactivity database" aside, and inline bash pipelines that duplicate functionality already shipped in scripts/. This is efficient with minor instances of over-explanation, not the fully lean anchor. | 4 / 5 |
Actionability | Every workflow is a complete, copy-paste-ready bash pipeline that handles URL-encoding and JSON parsing end-to-end, with specific example inputs covering the common cases (target search, bioactivity lookup, molecule properties, drug-likeness, interactions, adverse events, gene-disease). The bundled scripts (chembl_target.py, ro5_screen.py) are even more robust with error handling and rate limiting. | 5 / 5 |
Workflow Clarity | Five workflows are clearly numbered, titled, and self-contained, and the Reasoning Guidelines give a numbered analysis sequence (state values, apply rule sets, flag liabilities, suggest optimizations, cite source); inline code checks for empty results ("No interaction data found", "Target not found"). Minor validation gaps remain — no explicit feedback loop for API failures or rate-limit errors — and the cap at 3 does not apply since these are read-only single queries, not destructive or batch operations. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned and references/ADMET_REFERENCE.md is clearly signaled one level deep ("See references/ADMET_REFERENCE.md for detailed guidance"), but 2 of the 3 bundle files (scripts/chembl_target.py, scripts/ro5_screen.py) are never mentioned anywhere in the body, and ~150 lines of inline code duplicate them. Content that should point to the separate scripts is inlined and the navigation to most of the bundle is missing, which exceeds anchor 4's "minor organization gaps". | 3 / 5 |
Total | 16 / 20 Passed |