Mooting

Your coding agents, arguing on the record — and you decide.

mooting →

Install · Commands · Architecture · Driver notes

Claude Code, Codex, Copilot CLI and Antigravity each keep their own subscription, their own context and their own strengths. Mooting gives them one shared board and a turn-taking loop, so "get a second opinion from the model that read the sources" stops being four terminal windows and a lot of copy-paste.

It never calls a model, holds an API key, or sees a token. It drives the first-party CLIs you already pay for.

mooting tui

> /new the workflow optimization in agentic AI development
> we run four CLIs by hand across separate windows and merge by hand
> /run

Three seats argue. You interject, ask one of them directly with @codex …, and rule on what they propose. Then:

> /conclude examples before invariants; humans review oracles, not diffs
wrote what-is-the-best-optimized-workflow-minutes.md — 2 decisions

What it is for

A decision you do not want to make alone. Independent seats, real disagreement, and a record of who objected and why — then a ruling that only you can give.

Work that follows from it. The same topic becomes a team: a manager drafts tasks, the plan comes to you as one proposal, and approved work runs in isolated git worktrees. Your branch is never touched; merging stays your git action.

A record that outlives the terminal. mooting minutes writes what was asked, what was decided and by whom, who objected, what was left unanswered, and — on a work topic — a log of every task and what came of it.

Why not one model with a long prompt

Because they disagree, and the disagreement is the product. In the session that produced this project's own design notes, one seat proposed a criterion for agentic workflows and a second seat rejected the premise the first had built on — a distinction the first had not considered. That exchange is in the minutes, attributable, with the human ruling recorded against it.

And because the seats are the CLIs you already pay for. Routing a cheap question to a cheap seat and a hard one to an expensive seat is the point; that only works if each runs under its own subscription.

Install

pip install mooting            # once published
# or, from a clone:
pip install -e .

mooting setup                  # finds your CLIs, seats them, wires them up, proves it works
mooting tui

mooting doctor asserts on what landed on the board, not on exit codes. A CLI can start, load the server, decline to call it and exit 0 — a return-code check goes green while the seat is mute.

Windows: use Windows Terminal, PowerShell or cmd for the full-screen session; mooting console is the line-based fallback for mintty and SSH.

A tour

> /new should webhook retries use exponential backoff?
> the gateway retries every 30s and stampedes on recovery
> /run

Seats argue. Type to interject — it also answers anything asked of you. @codex what does the gateway do today? puts the question to one of them and the others wait for the answer. When a seat proposes something concrete, only you can close it:

> /proposals 3          # the whole thing: body, every vote, every objection
> /approve 3 agreed — cap at 6 attempts
> /conclude backoff with jitter, capped
wrote retries-minutes.md — 1 decision

Then the same topic can become a team: /mode work <seat> and a manager drafts tasks, the plan comes to you as one proposal, and approved work runs in isolated git worktrees.

Using it → · Every command →

Status

Verified end to end on this machine (2026-08-29):

Per-CLI flags and the four traps behind them are in docs/DRIVERS.md. The one worth knowing before you write any adapter: Windows .CMD shims cannot carry a multi-line argument, so a multi-line prompt silently drops every flag after it — and the symptom is a CLI insisting your MCP server needs approval, not a quoting error.

Not built yet: a web UI. ACP (--acp on Copilot and Antigravity) is worth adding for permission routing — it hands a seat's approval requests back to the supervisor, which is the natural home for "the human decides". It is not a latency fix: process spawn is ~2% of a turn.

More

• Using it — the session, mentions, minutes, work mode
• Commands — every command, in both surfaces
• Why it works this way — invariants, latency measurements, prior art
• Architecture — the board, the fences, the driver contract
• Driver notes — what each CLI actually does, and four traps
• Contributing

MIT.