Content
82%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 dense, well-organized instruction skill that earns its context budget with tool knowledge Claude cannot know (signatures, gotchas, thresholds). The main gaps are the absence of worked example invocations and explicit error-recovery checkpoints in the workflows.
Suggestions
Add one worked example invocation per workflow (e.g., PDBeSIFTS_get_best_structures(uniprot_id="P04637") followed by expected output fields) to make the guidance copy-paste ready.
Add explicit validation/recovery checkpoints to the workflows, such as what to do when BindingDB times out (60s+ noted in Limitations) or when no co-crystal structure exists for a target.
Move the Tool Inventory and Interpretation tables into a one-level-deep reference file (e.g., references/tools.md) to slim the always-loaded SKILL.md overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and dense throughout: "LOOK UP DON'T GUESS", "COMPUTE, DON'T DESCRIBE", terse tool signatures, a parameter-gotchas table, and domain-specific thresholds (resolution/R-free/pLDDT cutoffs, DoGSiteScorer >0.6) that Claude cannot be assumed to know. Not 4: there is essentially no over-explanation to trim — every section adds non-obvious domain or tool knowledge. | 5 / 5 |
Actionability | Concrete tool names with their parameters, a correct-vs-mistake gotchas table, artifact ligand filter lists, and numeric decision thresholds give mostly executable guidance. Not 5: workflows are phase outlines with no worked example invocation (e.g., an actual tool call with values) or sample output; not 3: the guidance is specific and directly executable, not pseudocode. | 4 / 5 |
Workflow Clarity | Three workflows with clearly sequenced phases and decision rules ("Resolution <2.5A for drug design", "Phase 4: PDBeValidation quality → binding site well-resolved?") plus evidence-grading tiers. Not 5: no explicit validate-and-retry feedback loops or error-recovery guidance (e.g., what to do when BindingDB times out or no co-crystal exists); not 3: checkpoints are present in each workflow, just implicit rather than looped. | 4 / 5 |
Progressive Disclosure | No bundle files exist, so all content is appropriately in the single SKILL.md, and it is well-organized into scannable sections (tool inventory, workflows, gotchas, evidence grading, interpretation, limitations). Not 5: the ~45-line tool inventory and interpretation tables are content that could be split into one-level-deep reference files to slim the always-loaded overview; not 3: nothing is buried and navigation is easy. | 4 / 5 |
Total | 17 / 20 Passed |