The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads highly sensitive user data including the Chrome browsing history database and synced device records from LevelDB. It copies these files to the /tmp directory for processing. While no network exfiltration occurs, the browsing history (URLs and page titles) is exposed to the AI agent.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs users to install the ccl_chromium_reader library directly from a GitHub repository to support synced mobile history parsing. This library is maintained by a known digital forensics organization.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted page titles and URLs from the user's history and includes them in the agent's context. A malicious website could potentially inject instructions via its page title which the agent might then follow during history analysis.
Ingestion points: chrome_history_query.py (reads from local Chrome databases)
Boundary markers: Absent (data is returned as a markdown list)
Capability inventory: File system read access to browser data
Sanitization: Partial (the script truncates titles to 75 characters)

chrome-history