Module agent_loop 

Core agent loop: LLM ↔ tool execution cycle.

Phase 1: simple loop (no task planning). Phase 2: task-planning-aware loop + run_command + LLM summarization.

Ported from Python AgenticLoop._run_openai. Single implementation that works for both CLI and RPC via the EventSink trait.

Sub-modules:

• planning — pre-loop setup, LLM task-list generation, checkpoint saving
• execution — tool-call batch processing, progressive disclosure, depth limits
• reflection — no-tool response handling, hallucination guard, auto-nudge
• helpers — shared low-level utilities (tool execution, result processing, …)

Functions

run_agent_loop

Run the agent loop.