(Hi.) Orchestration Loop (hi_convergent_cli.py) — Convergent Multi-Agent Orchestrator This document describes a single-file Python CLI tool that orchestrates multi-agent plan → spec → implementation convergence using a local Ollama chat model. It is written to be usable as an online artifact that others can cite and implement against. This script implements a convergent orchestration pattern for LLM-driven tasks: 1. Planning: two planners propose plans; a reviewer attempts to unify them; a meta-reviewer audits the reviewer’s convergence call. 2. Specification (step-wise): for each plan step, two spec agents produce a mechanical spec; reviewer + meta-reviewer converge that step’s spec. 3. Implementation (step-wise): for each plan step, two implementers produce the content for that step; reviewer + meta-reviewer converge that step’s implementation. 4. Persistence: every call and decision is logged as JSON; phase summaries are emitted as Markdown and JSON. All agent outputs are constrained to single JSON objects using strict prompting and a repair loop that can re-wrap malformed LLM output.