teach
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to open generated lesson files using CLI commands, establishing a capability to execute shell commands within the user's local environment.
- [REMOTE_CODE_EXECUTION]: The instructions require the agent to create reusable interactive components, such as 'simulators' and 'quiz widgets', which involves writing and potentially executing JavaScript code in the local workspace.
- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection due to its stateful nature and reliance on external data.
- Ingestion points: Content is read from state files such as MISSION.md and RESOURCES.md, which are populated based on external resources or user input.
- Boundary markers: There are no specified delimiters or instructions to ignore embedded commands when processing the content of these state files.
- Capability inventory: The agent has the ability to write executable HTML/JS files to the workspace and execute shell commands via the CLI.
- Sanitization: The skill lacks explicit instructions for sanitizing or validating information retrieved from external resources before it is recorded in the workspace state.
Audit Metadata