llm-council-harness
LLM Council Harness
Purpose
Implements a council whose members are headless coding agent CLIs — Claude Code, Codex CLI, Antigravity, Grok CLI — not just plain API calls. Each member is invoked as a subprocess, answers independently, and a boss model synthesizes. This is the implementation layer: given a council design is already wanted, this skill is how to actually run it against real CLI harnesses without hanging, leaking processes, or dying on one bad member.
When to use / when NOT to use
Use this skill when the user wants to run a council of coding harnesses in a script — "build a council of Claude Code and Codex", "run coding CLIs headless as council members", "harness deliberation script", "wire up claude -p and codex exec to deliberate" — including when they describe the mechanism without naming a council ("I want claude, codex and gemini CLIs to each answer the same question in a script and a boss model to pick").