screencapture
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses standard macOS system utilities, specifically
screencapturefor taking screenshots andosascriptfor querying window metadata. - [INDIRECT_PROMPT_INJECTION]: The agent ingests external visual data which could potentially contain adversarial content (e.g., text on screen attempting to redirect the agent's behavior).
- Ingestion points: Screenshots saved to the file system are read into the multimodal agent context via a Read tool.
- Boundary markers: No explicit delimiters are used for image data.
- Capability inventory: Access to
screencapture,osascript, andrmtools. - Sanitization: The skill enforces a strict cleanup policy, requiring the agent to delete the screenshot file immediately after reading.
Audit Metadata