teach
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The agent is tasked with reading and processing data from various local workspace files (MISSION.md, RESOURCES.md, assets, and learning records) as well as external online resources and communities. This creates a surface for indirect prompt injection, as malicious instructions within these sources could potentially influence the agent's behavior.
- Ingestion points: Local workspace files and external research content.
- Capability inventory: The agent writes local files (HTML lessons, Markdown records) and executes CLI commands.
- Boundary markers: The instructions do not define specific delimiters to help the agent distinguish external data from instructions.
- Sanitization: No sanitization or validation logic is prescribed for ingested content.
- [COMMAND_EXECUTION]: The instructions direct the agent to run CLI commands to open generated lesson files for the user. While intended as a convenience, this involves executing shell commands with parameters (file paths) determined by the agent's interaction with the workspace.
Audit Metadata