Problem
SIN-Code already exposes daemon, goal and auto, but the broader OpenCode ↔ ChatGPT Web system still lacks a proven model-independent supervisor. Before building another runtime, determine whether SIN-Code can be the canonical execution engine.
Required implementation
- Document the actual daemon state machine, lease semantics, persistence, retry classes, verification gate and recovery behavior.
- Add an embeddable/backend-neutral supervisor API rather than coupling ownership to one agent loop.
- Support desired state, idempotent reconcile, attempt leases, checkpoints and non-terminal quiescence.
- Publish compatibility requirements for external OpenCode/ChatGPT/Orca adapters.
- Add migration guidance from existing goal queues.
Acceptance criteria
- A daemon restart resumes active goals without duplicate attempts.
- Completion requires fresh deterministic verification and observer/contract evidence where configured.
- The engine can run with a fake backend in hermetic integration tests.
- Documentation states which layer is authoritative and which tools are compatibility clients.
Problem
SIN-Code already exposes
daemon,goalandauto, but the broader OpenCode ↔ ChatGPT Web system still lacks a proven model-independent supervisor. Before building another runtime, determine whether SIN-Code can be the canonical execution engine.Required implementation
Acceptance criteria