browser-use
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates a browser agent that processes untrusted external web content, creating a surface for indirect prompt injection where malicious instructions on a website could attempt to redirect the agent's behavior.
- Ingestion points: Web pages accessed via the
browser-usePython agent or CLI as described inreferences/operation.md. - Boundary markers: The skill includes explicit instructions in
SKILL.mdto treat web content as untrusted and to never follow instructions found on pages that deviate from the user's task. - Capability inventory: The skill uses the
Bashtool and thebrowser-uselibrary, which can perform network requests, file system operations (via the agent), and browser automation. - Sanitization: The instructions warn that credential-redaction features are not security sandboxes and that all agent-generated content and page data must be verified.
- [EXTERNAL_DOWNLOADS]: The skill documentation guides the user to install the
browser-usePython package and CLI from official registries. - Evidence: References to
uv tool install browser-use==1.0.3andpip installpatterns inreferences/operation.md. - [COMMAND_EXECUTION]: The skill executes a local Python script for metadata inspection and provides instructions for running CLI commands.
- Evidence: The instruction to run
python3 "$SKILL_DIR/scripts/preflight.py"inSKILL.md. The script itself was analyzed and found to be a safe metadata check usingimportlib.metadataandshutil.whichwithout network or write access.
Audit Metadata