pr-screenshot-docs

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
  • Data Exposure & Exfiltration (MEDIUM): The skill provides instructions to upload local files to an external, non-whitelisted hosting service.
  • Evidence: The workflow includes the command curl -F 'file=@screenshot.png' https://0x0.st in the 'Upload Screenshots' section.
  • Context: 0x0.st (The Null Pointer) is a public, anonymous, and ephemeral file hosting service. Uploading screenshots—which may contain sensitive UI data, internal URLs, or credentials visible in text fields—to such a service facilitates data exfiltration outside of organizational boundaries.
  • Indirect Prompt Injection (LOW): The skill involves processing and documenting UI components which are external, untrusted data sources.
  • Ingestion points: The skill utilizes browser_snapshot and browser_take_screenshot to ingest visual and structural data from web pages.
  • Boundary markers: Absent. There are no instructions provided to the agent to ignore or delimit potentially malicious instructions embedded in the UI elements being captured.
  • Capability inventory: The skill possesses network egress capabilities (curl) and browser navigation capabilities.
  • Sanitization: Absent. The captured data is processed directly into documentation without filtering or validation.
  • Command Execution (LOW): The skill suggests the execution of shell commands to perform network operations.
  • Evidence: The use of curl for file uploads.
  • Risk: While the commands are shown as examples, an autonomous agent following these instructions would execute external network requests that could bypass security egress policies.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 06:33 PM