tabbit-devtools

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXFILTRATION]: The skill reads the DevToolsActivePort file from Tabbit's user data directories. This file contains the sensitive debugging port and browser path required for remote control.
  • Evidence: read_active_port_file function in scripts/discover_tabbit_cdp.py searches %LOCALAPPDATA%, %APPDATA%, and ~/Library/Application Support for DevToolsActivePort.
  • [COMMAND_EXECUTION]: The skill executes external tools to perform browser automation.
  • Evidence: subprocess.run(command) in scripts/run_agent_browser_on_tabbit.py is used to launch the browser automation client.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes live web content through a browser, which creates a vulnerability surface for instructions embedded in web pages to influence the agent.
  • Evidence: The skill is designed to visit and inspect URLs using agent-browser (e.g., open <url> in SKILL.md).
  • [DYNAMIC_EXECUTION]: The skill uses dynamic loading to execute its internal helper scripts.
  • Evidence: runpy.run_path is used in scripts/read_current_tabbit_page.py to execute discover_tabbit_cdp.py.
  • [EXTERNAL_DOWNLOADS]: The skill may download the agent-browser package from a public registry if it is not already installed.
  • Evidence: npx --yes agent-browser is used as a fallback execution method in scripts/run_agent_browser_on_tabbit.py.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 04:01 PM
Security Audit — agent-trust-hub — tabbit-devtools