The broad-bound listener and agent routes appear to lack authentication, while the permission route can grant tools; this is a serious exposure when the server is reachable from an untrusted network. The behavior aligns with the package’s server purpose, so the evidence supports a critical vulnerability warning rather than a malicious verdict.
Static reason
No blocking static signals were detected.; source matched previously finalized malicious package; routed for review
Trigger
Starting the server with the documented server command or the cortex-server/cortex-daemon binary; the listener uses PORT (default 4000).
Impact
If the host is reachable from an untrusted network, a remote caller may cause agent actions or change tool permissions using the server process's access to configured services and project resources.
Mechanism
A network caller can submit agent requests to /v1/messages and ask /permissions/tool/:name to grant a tool by registering a whitelist policy.
Evidence
dist/index.jsdist/routes/messages.jsdist/routes/permissions.js
Network endpoints3
0.0.0.0:<PORT>/v1/messages