The package has a real, user-activated remote agent capability risk: the headless default combines broad network exposure with automatic tool approval. This supports a warning, while the inspected source does not establish a malicious install hook or credential attack.
Static reason
No blocking static signals were detected.
Trigger
A user starts the server or daemon in a non-interactive environment and a remote client submits a messages request.
Impact
A remote client may induce model-backed tool actions against the server's configured project without an interactive approval prompt; the exact actions depend on available tools and model behavior.
Mechanism
The server accepts messages requests without an authentication check in the inspected route, defaults to built-in tools, and uses automatic permission mode in headless operation unless the opt-out is set.
Evidence
dist/index.jsdist/routes/messages.js