web-access
Warn
Audited by Gen Agent Trust Hub on Jun 14, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [DATA_EXFILTRATION]: The
scripts/find-url.mjsscript accesses and queries local browser history and bookmarks databases for Chrome and Edge. This allows the AI agent to retrieve sensitive information about the user's browsing habits, frequently visited sites, and internal organizational URLs. - [REMOTE_CODE_EXECUTION]: Through the
scripts/cdp-proxy.mjscomponent, the skill exposes an API that allows the agent to execute arbitrary JavaScript code (/eval) in a connected browser tab. This capability can be used to interact with web pages using the user's active login sessions, enabling data extraction or actions within authenticated environments. - [COMMAND_EXECUTION]: The skill executes local system commands to perform its operations. Specifically,
scripts/find-url.mjsruns thesqlite3utility to query browser databases, andscripts/check-deps.mjsmanages the background execution of the browser proxy service usingnode. - [PROMPT_INJECTION]: The skill's core functionality involving the ingestion of content from arbitrary web pages (via CDP,
curl, orWebFetch) creates a surface for Indirect Prompt Injection attacks. - Ingestion points: External web content retrieved from URLs provided by the user or discovered during navigation.
- Boundary markers: The skill does not implement explicit delimiters or
Audit Metadata