ncbi-pmc-skill
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection Surface: The skill fetches XML content from the NCBI PMC API and processes it into summaries. This creates a surface where instructions embedded in external research data could potentially influence the agent.
- Ingestion points: Data enters via the requests.get call in scripts/ncbi_pmc.py.
- Boundary markers: The skill does not currently use specific delimiters or warnings to isolate external content from the agent's instructions.
- Capability inventory: The skill possesses network GET capabilities and the ability to write files to the local file system.
- Sanitization: Content is truncated via the _compact function, but no semantic filtering or instruction-detection is applied.
- File System Write Access: The save_raw and raw_output_path parameters allow the skill to write API responses to a user-defined path. If the execution environment does not enforce strict path sandboxing, this could potentially be used to overwrite files in the local environment.
- Credential Handling: The skill appropriately manages API keys by retrieving them from environment variables (NCBI_API_KEY, NCBI_EUTILS_API_KEY), which follows best practices for secret management.
Audit Metadata