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 API content is genuinely strong — executable examples for every major workflow, clear result-access patterns, and a well-signaled set of one-level-deep references. It is dragged down by substantial generated template boilerplate (When to Use, Key Features, Example Usage, Implementation Details, Quick Validation, K-Dense promotion) and a generic Recommended Workflow section that provides no concrete Rowan-specific sequence or batch validation steps.
Suggestions
Delete the template boilerplate sections (When to Use, Key Features, Example Usage, Implementation Details, Quick Validation) that restate the description or contain no-op filler, and remove the unrelated K-Dense Web promotional section.
Replace the generic Recommended Workflow / Quick Validation filler with the real Rowan sequence: set ROWAN_API_KEY -> rowan.whoami() to verify credits -> submit workflow -> wait_for_result(timeout) -> check workflow.status -> fetch_latest(in_place=True) -> read workflow.data, including an explicit failure/retry branch and a validation step for batch submissions.
Fix Pattern 2 (Virtual Screening) so it is copy-paste runnable: define the pocket center coordinates and compound_library, and add "import stjames" to the docking code snippet.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Roughly 100 of the ~527 lines are generated template boilerplate: "Use this skill when the request matches its documented task boundary", a Key Features section that repeats the frontmatter description verbatim, an Example Usage artifact ("Skill directory: 20260316/scientific-skills/Data Analytics/rowan / No packaged executable script was detected"), a self-referential placeholder ("See `## Overview` above" — Overview actually appears later in the file), "No local script validation step is required for this skill", and an unrelated promotional section recommending K-Dense Web. This matches the 'noticeably verbose; several unnecessary explanations or padded sections' anchor; it is below 3 because the padding is substantial boilerplate rather than occasional over-explanation, and above 1 because the core API content is dense, assumes competence, and never explains concepts Claude already knows. | 2 / 5 |
Actionability | The body provides concrete, executable Python for installation, authentication verification, five core workflows (pKa, conformer search, optimization, docking, cofolding), the RDKit-native API, and error handling with a timeout and status checks. Minor gaps keep it from a 5: Pattern 2 uses an undefined "compound_library" variable and a literal placeholder pocket ("center": [x, y, z]), and the docking example calls stjames.Molecule without importing stjames in that snippet. It sits above anchor 3 because the code is genuinely executable, not pseudocode. | 4 / 5 |
Workflow Clarity | Each core workflow demonstrates a consistent, readable sequence — submit, workflow.wait_for_result(), workflow.fetch_latest(in_place=True), then access workflow.data — and the Error Handling section adds status checks with a timeout and failure branch. However, the "Recommended Workflow" section is generic filler ("Validate the request against the skill boundary... Select the documented execution path") with no concrete commands, and the batch operations (batch_submit_workflow, batch_poll_status) have no validation or retry feedback loops, which the rubric caps at 3. | 3 / 5 |
Progressive Disclosure | Six real reference files (api_reference.md, workflow_types.md, rdkit_native.md, molecule_handling.md, proteins_and_organization.md, results_interpretation.md) are all listed with per-file descriptions in the Reference Documentation section and are one level deep (no reference points to further reference files). It falls short of a 5 because a fair amount of detailed API material (workflow management, computational method lists, RDKit function inventory) is inlined in SKILL.md instead of being pushed to those files, and rdkit_native.md is referenced mid-body rather than only from the reference index. | 4 / 5 |
Total | 13 / 20 Passed |