Content
92%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A concise, highly actionable orchestration skill with a clear phased workflow and built-in quality tiers. Its main gap is progressive disclosure: everything lives inline in one file with no reference structure for the sizable tool/evidence detail.
Suggestions
Move the Tool Parameter Reference and Evidence Grading tables into a short REFERENCE.md and link to it from the body so the main SKILL.md stays a lean overview.
Add one concrete Python example under "COMPUTE, DON'T DESCRIBE" (e.g. grouping PTM sites by modification type with pandas) to back the directive with executable guidance.
Add an explicit validate-and-retry note for the compute step (e.g. re-query iPTMnet if a UniProt accession returns empty) to make the workflow's error-recovery loop explicit.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean bullet/table format with no padding or explanations of concepts Claude already knows; directives like "LOOK UP DON'T GUESS" and "COMPUTE, DON'T DESCRIBE" assume competence and every token earns its place. | 3 / 3 |
Actionability | Concrete, copy-paste-ready tool calls with real parameters (e.g. `iPTMnet_get_ptm_sites(operation="get_ptm_sites", uniprot_id="P04637")`) plus a Tool Parameter Reference table and fallback matrix. | 3 / 3 |
Workflow Clarity | Clearly sequenced Phase 0→5 pipeline with a disambiguation checkpoint in Phase 0 ("Select human entry if multiple hits") and a tiered Evidence Grading table that functions as a quality checklist for synthesis. | 3 / 3 |
Progressive Disclosure | Content is well-organized into labeled sections but the skill is a single monolithic file over 50 lines with no external references or one-level-deep navigation to split out the tool catalog or evidence-grading detail. | 2 / 3 |
Total | 11 / 12 Passed |