Server has 4 tools with basic descriptions and minimal schemas. All tools lack parameter type definitions in visible schema (only string descriptions provided). Descriptions are present but generic (avg ~80 chars, below 194-char baseline). No output schemas documented. Error handling returns plain strings without recovery guidance. Tool names lack clear verb-noun structure ('parse_and_execute' violates single-responsibility principle). No input validation, no enum constraints, no pagination support. Security concern: parse_and_execute accepts arbitrary ChimeraX commands with no sanitization, risking command injection.
Fetch an AlphaFold structure by UniProt ID.
Predict the structure of a protein sequence using AlphaFold.
Execute a ChimeraX command directly.
Search for protein PDBs based on the specified type.
parse_and_execute violates single-responsibility principle: name contains 'and', tool accepts arbitrary ChimeraX commands without validation or sanitization, risking command injection attacks.
No input schemas with type definitions visible. All parameters documented only as strings in descriptions, not in formal JSON Schema. Rubric requires type definitions for schema score >30.
No output schemas documented. Tools return plain strings; LLMs cannot parse structured results or chain downstream calls. Baseline: 100% of A+ tools have documented return types.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | <=2025-11-25 | v2 |
Error handling returns plain error strings without recovery guidance. Rubric requires: 'Error responses must tell the LLM what to do next.' Current: 'Error: {error}' gives agent nothing actionable.
search_protein_pdbs accepts num_results with no bounds validation. Unbounded integers let LLMs pass absurd values (e.g. 1000000) that break APIs or cause timeouts.