peptide-library-research
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches research data, study abstracts, and regulatory information from the vendor's domains at
peptidelibrary.appandapi.peptidelibrary.app. These network operations are intrinsic to the skill's purpose as a research tool and target official vendor infrastructure. - [INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes external content from peptide profiles and research articles, which constitutes a surface for indirect prompt injection.
- Ingestion points: Data enters the agent's context through MCP tool outputs (e.g.,
get_peptide_details,get_human_study), REST API endpoints, and library pages. - Boundary markers: The skill provides a structured workflow requiring the agent to categorize findings into specific evidence tiers (e.g., controlled human trials, animal research) and explicitly cite primary sources, which helps delineate external data from agent responses.
- Capability inventory: The skill is restricted to read-only research tools and basic lab arithmetic utilities; it possesses no capabilities for file system modification, command execution, or non-vendor network communication.
- Sanitization: Instructions mandate the reporting of documented limitations (such as bias or small sample sizes) and prohibit the agent from filling gaps in data, ensuring a high-fidelity representation of the source material.
- [SAFE]: The instructions include robust safety protocols that strictly forbid the generation of medical advice, personal dosing protocols, reconstitution instructions, or vendor recommendations. These constraints effectively scope the skill to educational and research use.
Audit Metadata