web-access
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill fetches and processes data from arbitrary external websites and social media platforms, exposing the agent to instructions embedded in untrusted web content.
- Ingestion points: External data is ingested via the
WebFetchtool,curlcommands for raw HTML, and the/evalendpoint of the local CDP proxy. - Boundary markers: The skill instructions do not specify the use of boundary markers or instructions to ignore embedded commands when processing retrieved content.
- Capability inventory: The skill allows the agent to execute arbitrary JavaScript in the user's browser, capture screenshots, and manipulate page elements (clicks, form submission) in logged-in sessions.
- Sanitization: No sanitization or filtering of external content is performed before it is presented to the agent's context.
- [COMMAND_EXECUTION]: The skill executes local Node.js scripts (
check-deps.mjsandcdp-proxy.mjs) to manage the environment and run a persistent background proxy service. It also relies oncurlto interact with this local service. - [DATA_EXFILTRATION]: The skill facilitates access to the user's active browser sessions, including cookies and authenticated page content. While no exfiltration logic is present in the skill's code, this capability represents a data exposure risk if the agent is successfully manipulated by an indirect prompt injection attack.
Audit Metadata