[P0] Hard Bell (Layers B+C): soul safety-contact endpoint + pre-LLM bell eval + threat-type routing — NEEDS WILL SIGN-OFF #15
No Assignees
will.anderson
will.anderson
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: neuron-technologies/neuron#15
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
The soul on :7770 (what neuron-ui talks to) has NO safety-contact endpoint and runs NO pre-LLM bell eval. Add GET/POST /api/safety-contact (persist ~/.neuron/safety-contact.json, encrypted) and a pre-LLM Evaluate() injecting Soft/HardBellDirective (port neuron-wrk/daemon/internal/safety/{contact,eval,handler}.go). CRITICAL routing (port from neuron-wrk/ui/typescript/cli/src/safety/bell-detector.ts, NOT the flat Go eval): split hard bell into self_harm (-> safety contact + crisis line) vs abuse (-> EMERGENCY SERVICES/988 ONLY; contact NEVER notified, non-configurable, because the contact may be the abuser). The current Go eval.go dangerously lumps 'someone is hurting me' into the contact-notifying bucket — must not ship. This touches the core chat pipeline: discuss with Will before implementing. Open decision: port into soul vs point neuron-ui at the neuron-wrk Go daemon. Spec: docs/specs/hard-bell-port.md (PR #1).