atlas

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.70). This skill reads and queries the user's Atlas profile data (History SQLite, Bookmarks JSON) and enumerates open tab titles/URLs via AppleScript (see scripts/atlas_common.py get_history_path/get_bookmarks_path and scripts/atlas_cli.py get_tabs/search_history/get_bookmarks), which are local copies of arbitrary public third‑party webpages and can be used to drive actions (open/focus/close tabs), so untrusted web content could indirectly influence agent behavior.