Content
63%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 research skill with a clear phased workflow, concrete tool-call examples, fallback chains, and checklists. Its main weaknesses are token efficiency (duplicated tool details and basic pharmacology definitions) and the absence of any progressive disclosure — everything lives in one long SKILL.md.
Suggestions
Remove or compress content Claude already knows (the agonist/antagonist/partial-agonist definitions in Phase 2 and the clinical biased-agonism example in the opening paragraph), keeping only the skill-specific instruction to always classify ligands by pharmacological type.
Deduplicate the 'Tool Parameter Reference' table against the Phase 1-5 code examples — keep the table for parameter formats and trim the repeated per-call commentary, or move the full table plus the GPCRdb entry-name catalogue into a references/ file that SKILL.md points to.
Add a short error-recovery loop for the common failure mode (no structures found for a receptor / wrong entry name), e.g., re-run GPCRdb_list_proteins, then fall back to RCSB PDB search, and document the `limit` parameter of SAbDab_search_structures in the parameter table since it is used in Phase 5.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~290-line body is mostly efficient but includes explanations Claude already knows ('Agonist: activates receptor', 'Antagonist/Inverse agonist: blocks or suppresses receptor', the opening paragraph restating biased-agonism pharmacology) and the 'Tool Parameter Reference' table duplicates parameter details shown again in the Phase 1-5 code blocks. Not 4 because the duplication and basic pharmacology definitions are more than minor trimmable instances. | 3 / 5 |
Actionability | Mostly executable guidance: concrete tool calls with real parameters ('GPCRdb_get_structures(protein="adrb2_human", state="inactive")'), real PDB IDs ('2rh1', '6W41'), BSA thresholds (>1500 Ų strong), and worked patterns. Not 5 because cross-references to other skills' tools ('PubChem_get_CID_by_compound_name', 'ChEMBL_search_molecules') are invoked without enough detail to execute, and 'SAbDab_search_structures(query="EGFR", limit=20)' uses a `limit` parameter not documented in the parameter table. | 4 / 5 |
Workflow Clarity | A clear 5-phase sequence with fallback chains for failed lookups and per-task completeness checklists; these are read-only research operations, so the destructive/batch validation cap does not apply. Not 5 because there are no explicit feedback loops (e.g., what to do when GPCRdb_get_structures returns no entries for a receptor beyond the one-line fallback table). | 4 / 5 |
Progressive Disclosure | Good section structure with clear headers, but the skill is a monolithic ~290-line file: the full tool parameter reference table, entry-name catalogue, and per-phase call examples are inlined content that naturally belongs in a references/ file. No bundle files (references/, scripts/, assets/) exist, and the 'Key References' section only lists external URLs. Not 4 because the inline API reference is substantial enough that the file would clearly benefit from splitting. | 3 / 5 |
Total | 14 / 20 Passed |