osascript-chrome

Warn

Audited by Gen Agent Trust Hub on Jun 17, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The skill contains tools specifically designed to extract session-sensitive information from active browser tabs. The tab-cookies.py script retrieves non-HttpOnly cookies, and tab-storage.py is documented to extract authentication tokens from localStorage and sessionStorage.
  • [DATA_EXFILTRATION]: Multiple scripts access and export private browser profile data from the local filesystem. chrome-history.py, chrome-downloads.py, and chrome-bookmarks.py read from the Chrome profile directory (e.g., ~/Library/Application Support/Google/Chrome/), exposing the user's full browsing history, download logs, and bookmark collections.
  • [COMMAND_EXECUTION]: The skill utilizes subprocess.run to execute dynamically generated JXA (JavaScript for Automation) scripts via the osascript utility. This allows the skill to programmatically control browser windows, tabs, and page navigation.
  • [COMMAND_EXECUTION]: The scripts execute-js-in-tab.py and read-tab.py allow the agent to execute arbitrary JavaScript code within the context of any matching open browser tab. This provides an unconstrained mechanism for interacting with web applications and extracting any data visible in the DOM.
  • [PROMPT_INJECTION]: The skill facilitates indirect prompt injection by scraping content from external websites and returning it to the agent's context without sanitization.
  • Ingestion points: scripts/scrape-page.py, scripts/scrape-links.py, and scripts/read-tab.py all ingest content from external web pages via JavaScript execution.
  • Boundary markers: None identified; extracted text, headers, and metadata are returned directly to the agent without delimiters or instructions to ignore embedded commands.
  • Capability inventory: Extensive browser control including automated form filling (form-fill.py), tab management, and reading of local profile data (history, cookies).
  • Sanitization: None; the scripts perform string slicing (e.g., .slice(0, 5000)) for length but do not sanitize or filter the content for potentially malicious instructions.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 17, 2026, 09:45 AM
Security Audit — agent-trust-hub — osascript-chrome