fast-browser-use

Pass

Audited by Gen Agent Trust Hub on Nov 22, 2026

Risk Level: SAFEDATA_EXFILTRATIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill provides tools that allow an agent to extract potentially sensitive session information from the browser context.
  • src/tools/cookies.rs: The GetCookiesTool enables the agent to retrieve all cookies from the current browser session, which may include session identifiers or authentication tokens.
  • src/tools/local_storage.rs: The GetLocalStorageTool allows for the extraction of all data stored in the browser's local storage for the current origin.
  • [DYNAMIC_EXECUTION]: The skill includes an evaluate tool that permits the execution of arbitrary JavaScript code within the browser.
  • src/tools/evaluate.rs: The EvaluateTool executes code provided via parameters using the tab.evaluate method. While a standard feature for browser automation, this capability can be leveraged for unauthorized actions or to facilitate data exfiltration if the agent is directed to malicious sites.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted web data and possesses powerful capabilities, creating a significant attack surface for indirect prompt injection.
  • Ingestion points: Untrusted data from external websites enters the agent's context through tools such as navigate, get_markdown, and snapshot (found in src/tools/navigate.rs, src/tools/markdown.rs, and src/tools/snapshot.rs).
  • Boundary markers: The tool outputs lack explicit delimiters or instructions for the agent to ignore potentially malicious content embedded in the extracted web data.
  • Capability inventory: The skill provides access to arbitrary JavaScript execution (evaluate), file system writes (screenshot, annotate), and network operations through the browser.
  • Sanitization: No mechanisms for sanitizing or filtering the extracted content are implemented to prevent embedded instructions from influencing subsequent agent actions.
  • [COMMAND_EXECUTION]: The skill manages the lifecycle of an external browser process.
  • src/browser/session.rs: The launch method spawns a Chrome or Chromium process using the headless_chrome crate, utilizing paths that can be configured via environment variables.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 22, 2026, 01:11 AM