Content
70%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 well-structured, highly actionable reference for GPCRdb/SAbDab/PDBePISA tooling with clear phased workflows and validation gates. Its weaknesses are inlined background pharmacology explanations and a monolithic single-file layout that would benefit from splitting the tool reference and entry-name catalog into reference files.
Suggestions
Remove definitional padding Claude already knows (agonist/antagonist/partial-agonist glosses, the CDR-H3 variability note, and the biased-agonism clinical rationale paragraph) and keep only the tool-specific interpretation guidance like BSA thresholds and Ballesteros-Weinstein numbering conventions.
Split the Tool Parameter Reference table, GPCRdb entry-name catalog, and Common Research Patterns into separate reference files under references/ (e.g., references/tool-reference.md, references/patterns.md) and point to them one level deep from SKILL.md.
Reconcile the parameter documentation — either add limit (and any other optional parameters) to the SAbDab_search_structures row in the tool reference table or drop it from the Phase 5 example — and parameterize the cross-references to PubChem/ChEMBL tools so every call shown is executable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body spends tokens defining concepts Claude already knows — "Agonist: activates receptor", "Antagonist/Inverse agonist: blocks or suppresses receptor", "Partial agonist: submaximal activation", and "CDR-H3 is most variable and typically dominates antigen contact" — plus a lengthy opening paragraph on biased-agonism clinical rationale. It is above 2 because the bulk of the content (tool tables, phase code, patterns) is genuinely non-obvious API knowledge, but the explanatory padding keeps it below 4. | 3 / 5 |
Actionability | Concrete executable calls with real argument values (GPCRdb_list_proteins(family="adrenoceptors"), PDBePISA_get_interfaces(pdb_id="2rh1")) plus documented return values and fallback chains give copy-paste-ready guidance. Not 5 because of small inconsistencies — SAbDab_search_structures(query="EGFR", limit=20) uses a limit parameter the tool reference table does not list — and cross-references to PubChem/ChEMBL tools that are only named, not parameterized. | 4 / 5 |
Workflow Clarity | Five phases are clearly sequenced with a workflow overview, four Input/Flow/Output research patterns, an explicit verification gate ("LOOK UP DON'T GUESS: never assume GPCRdb entry names... always use GPCRdb_list_proteins"), fallback chains for error recovery, and per-domain completeness checklists. This matches the anchor-5 package of clear sequence, validation steps, feedback loops, and checklists. | 5 / 5 |
Progressive Disclosure | The skill is a single 290-line SKILL.md with no bundle files — the full tool parameter reference, the GPCRdb entry-name catalog, and four research patterns are inlined where reference files would keep the overview lean. Section headers give it real structure (above 2), but content that clearly belongs in separate files is inline and there are no external references at all, which fits the anchor-3 example of structure with inlined content. | 3 / 5 |
Total | 15 / 20 Passed |