[ENH] Decompose CNC into Planner and Executor modules

[OppaAI]

Summary

CNC currently acts as orchestrator, context builder, stream handler, and memory coordinator all in one. As GRACE grows more capable this becomes a bottleneck. Decomposing CNC into specialized modules makes the architecture more scalable and maintainable.

Current Behavior

CNC does everything:

• Receives user input (orchestrator)
• Builds context (context builder)
• Calls Cosmos (inference handler)
• Streams response (stream handler)
• Manages memory via MCC (memory coordinator)

Proposed Enhancement

Decompose into specialized modules:

CNC (conductor — thin orchestrator only)
    ├── Planner    — decides what to do and how
    │   ├── context assembly
    │   ├── tool selection (future agentic)
    │   └── goal tracking
    └── Executor   — carries out the plan
        ├── Cosmos inference
        ├── streaming
        └── memory updates

# Future CNC._handle()
plan = await self.planner.plan(user_input, memory_context)
result = await self.executor.execute(plan)
await self.mcc.store(result)

Impact

• CNC becomes a thin conductor, not the whole orchestra
• Each module has single responsibility
• Easier to add agentic capabilities (Planner decides tools)
• Easier to test each module independently
• Maps to human prefrontal cortex (planner) + motor cortex (executor)

Notes

• Not needed for M1-M3 — implement when agentic missions begin (M10/M11)
• Planner module is the foundation for MCP tool server
• CNC decomposition aligns with AIVA server taking over orchestration role