agent-browser

Pass

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the agent-browser CLI to perform various browser actions such as navigating URLs, clicking elements, and filling forms.\n- [DATA_EXFILTRATION]: The skill can extract data from web pages using snapshots, text retrieval, and screenshots. This data is returned to the agent context or saved to local files.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its processing of untrusted web content.\n
  • Ingestion points: Untrusted data enters the agent context via agent-browser snapshot and agent-browser get commands.\n
  • Boundary markers: The skill documentation does not mention the use of delimiters or instructions to ignore embedded commands within the ingested web data.\n
  • Capability inventory: The skill provides significant interaction capabilities, including navigating, clicking, and filling inputs, which could be abused by malicious instructions in a web page.\n
  • Sanitization: No sanitization or validation of the retrieved web content is indicated.\n- [CREDENTIALS_UNSAFE]: The skill includes commands to save and load browser session state (e.g., agent-browser state save auth.json), which stores sensitive cookies and authentication information in local files.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 29, 2026, 08:21 PM