tooluniverse-natural-product-dereplication

Pass

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and external data from chemical databases. Ingestion points: User-supplied chemical identifiers (names, formulas, SMILES) and API responses from NPAtlas, ClassyFire, and PubChem. Boundary markers: The skill instructions emphasize verification but do not include explicit prompt/data delimiters. Capability inventory: Limited to specific chemical database queries; no access to sensitive files, credentials, or arbitrary network endpoints. Sanitization: Input is processed by chemical name/structure parsers, which inherently filter non-chemical data.
  • [SAFE]: Analysis of the skill instructions and tools revealed no malicious patterns. The skill focuses on legitimate chemical informatics by querying trusted databases. No evidence of prompt injection, data exfiltration, or obfuscation was found.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 7, 2026, 01:50 PM