skills/openclaw/skills/agent-browser/Gen Agent Trust Hub

agent-browser

Pass

Audited by Gen Agent Trust Hub on Mar 27, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches the agent-browser package and associated Chromium binaries from Vercel Labs' official repositories and well-known package registries.
  • [DATA_EXFILTRATION]: Provides capabilities to retrieve and persist sensitive session data for legitimate session management. Evidence: Includes commands to export authentication cookies and local storage state to local JSON files (e.g., agent-browser state save auth.json).
  • [PROMPT_INJECTION]: Contains a surface for indirect prompt injection from processing external web data. 1. Ingestion points: Untrusted web content via agent-browser open and snapshot in SKILL.md. 2. Boundary markers: Absent. 3. Capability inventory: CLI subprocess execution, file system writes for screenshots/state, and network navigation in SKILL.md. 4. Sanitization: Absent.
  • [COMMAND_EXECUTION]: Orchestrates browser automation tasks through the agent-browser CLI, allowing for deterministic element selection and multi-step workflow automation.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 27, 2026, 09:01 AM