Screenshot Analysis
Pass
Audited by Gen Agent Trust Hub on Mar 18, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It instructs the AI to analyze screenshots which may contain untrusted text content from web pages or applications. An attacker could place malicious text within an image to trick the AI into executing unauthorized actions.\n
- Ingestion points: The
analyze_screenshottool inSKILL.mdtakes a screenshot for analysis, which serves as an untrusted external data source.\n - Boundary markers: Absent. The prompt templates provided in the skill do not include instructions for the AI to ignore or treat text-based commands within the visual data as untrusted.\n
- Capability inventory: The skill leverages vision capabilities (OpenAI Vision API) to interpret visual and textual data from images.\n
- Sanitization: Not mentioned. No process is described for filtering or validating the textual content extracted from screenshots before it is processed by the model.\n- [COMMAND_EXECUTION]: The
analyze_screenshottool accepts ascreenshotparameter defined as a full file path. While intended for image files, this interface provides the agent with file system access capabilities that could be misused to probe for or access sensitive files if the hosting environment is not properly sandboxed.
Audit Metadata