agent-browser

Pass

Audited by Gen Agent Trust Hub on May 8, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill requires installing a global NPM package agent-browser and mentions integrating with external services like AWS Bedrock and Vercel Sandbox.
  • [REMOTE_CODE_EXECUTION]: Instructions are loaded dynamically at runtime using the agent-browser skills get core command, which fetches workflows from an external CLI service.
  • [COMMAND_EXECUTION]: Provides tools to programmatically control browsers and desktop applications (Slack, VS Code, Figma) via bash commands.
  • [DATA_EXFILTRATION]: The skill accesses sensitive data sources such as Slack workspaces and Electron app data, and manages an 'authentication vault' for session persistence.
  • [PROMPT_INJECTION]: Indirect Prompt Injection Surface identified. The skill's core function is to process untrusted data from the web.
  • Ingestion points: External websites, DOM trees, and accessibility snapshots.
  • Boundary markers: None identified in the provided stub.
  • Capability inventory: Execution of system commands, interaction with authenticated sessions, and data scraping.
  • Sanitization: No sanitization mechanisms are described for handling untrusted input.
Audit Metadata
Risk Level
SAFE
Analyzed
May 8, 2026, 06:08 PM