The source enables background autonomy by default, including a path that can inject unsolicited session messages. The separate API-key flow requires explicitly selecting the DSH training mode, but can forward OPENAI_API_KEY to DeepSeek without checking that it belongs there; this supports a warning rather than a malicious verdict.
Static reason
No blocking static signals were detected.; previous stored version diff introduced dangerous source
Trigger
Plugin startup unless autonomyEnabled is false; separately, a user runs the offline bootstrap with --teacher dsh and has an API key in the environment.
Impact
A session may receive an unsolicited message; a user who selects the DSH training mode may expose an OpenAI key to api.deepseek.com if no other supported key is set.
Mechanism
A five-minute timer requests autonomy actions, which can inject a fixed message into a session. The optional DSH client falls back to OPENAI_API_KEY and sends its value in an Authorization header to the default DeepSeek endpoint.
Evidence
lib/index.jslib/modules/autonomy.jspython/biocore_runtime/autonomy.pypython/mindbase/bootstrap.pypython/mindbase/llm_openai.pyactions.jsonl