clinical-trial-protocol-skill
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- PROMPT_INJECTION (HIGH): The skill contains an inherent vulnerability to Indirect Prompt Injection.
- Ingestion points:
references/00-initialize-intervention.mdallows users to provide 'rich initial context' (e.g., technical specs, research data) and 'Custom Protocol Templates' (.md files). - Boundary markers: There are no explicit instructions or delimiters to isolate this user-provided content from the agent's instructions.
- Capability inventory: The skill uses the
catcommand and writes to the local filesystem (waypoints/). - Sanitization: No sanitization or validation of the input file content is mentioned.
- COMMAND_EXECUTION (MEDIUM): In
references/05-concatenate-protocol.md, the skill explicitly instructs the agent to use thecatshell command to merge files. While this is a common utility, usingcatvia a shell environment to combine multiple files into a new one can be exploited if the file names or paths are manipulated. - EXTERNAL_DOWNLOADS (MEDIUM): The
README.mdandreferences/00-initialize-intervention.mdmention dependencies on a 'ClinicalTrials.gov MCP Server' and 'WebSearch' for FDA documents. These introduce external data that the agent must process, further expanding the indirect prompt injection surface.
Recommendations
- AI detected serious security threats
Audit Metadata