Plugin OpenClaw Ecosystem

Agent Meeting

Structured multi-agent deliberation — decisions that scale without burning your token budget

Overview

Agent Meeting orchestrates any number of OpenClaw agents through a structured turn-based discussion using the Mediator Pattern. A coordinator routes all messages, agents embed stance markers in every response, and history is compressed to ~500 tokens per round — so meetings stay sharp no matter how long they run.

Capabilities

Key Features

Mediator Pattern

A single coordinator routes every message — eliminating the N×N chatter explosion that kills naïve multi-agent setups. Agents only ever talk to the coordinator, never directly to each other.

Rolling Summary

After every round, the full transcript is compressed to ~500 tokens. Each subsequent round receives only the summary plus the new question — keeping token usage O(1) per round regardless of meeting length.

Stance Detection & Consensus

Every agent embeds [STANCE: AGREE / DISAGREE / NEUTRAL] in its response. The coordinator parses stances deterministically and triggers early exit the moment full or majority consensus is reached.

Auto Minutes

When the meeting ends — by consensus or by exhausting max rounds — a Markdown minutes file is written automatically: topic, participants, per-round transcripts, rolling summaries, and the final consensus verdict.

Early Exit

No need to sit through five rounds if agents agree in two. Meetings terminate automatically on full consensus (all agree) or majority consensus (≥ 2/3 agree, 0 disagree).

CLI Interface

Two commands cover every use case. `meeting quick` runs a 3-round session with auto output. `meeting start` gives full control over topic, agents, rounds, and output path.

Architecture

How It Works

Each round follows a deterministic six-step cycle. The coordinator drives every step — agents are stateless responders that receive the rolling summary and the current question, nothing more.

1. Coordinator broadcasts Sends rolling summary + round question to all agents in parallel

2. Agents respond Each agent replies with reasoning and a [STANCE: AGREE/DISAGREE/NEUTRAL] marker

3. Stance extraction Coordinator parses stance tokens deterministically — no LLM interpretation

4. Consensus check If full or majority consensus: write minutes and exit early

5. Compress history Summarizer compresses the full transcript to ~500 tokens for the next round

6. Repeat or finish Cycle continues until consensus or max rounds; minutes are always written

O(1)

Token growth per round

Stance types (AGREE / DISAGREE / NEUTRAL)

Auto

Early exit on consensus

1 cmd

To start any meeting

Usage

Two Commands, Any Meeting

From a quick 3-round sanity check to a fully parameterised strategy session — it's one command either way.

Quick Meeting

3 rounds

Spin up a rapid discussion with any two agents. Output goes to ./minutes.md automatically.

$ meeting quick "TypeScript or JavaScript?" researcher coding

Structured Meeting

Full control

Specify agents, max rounds, and output path. Works with any OpenClaw agent IDs you have configured.

$ meeting start \
  --topic "Q2 priority: feature A vs B" \
  --agents researcher,coding,content \
  --rounds 5 \
  --output ./minutes.md

Use Cases

What Gets Decided

Any question that benefits from multiple expert perspectives — without the coordination overhead of a human meeting.

Architecture Decisions

Debate monolith vs microservices, database choices, service boundaries, and API contract design with your researcher and coding agents.

System Design Trade-offs

Tech Stack Choices

Let specialist agents weigh in on language, framework, and toolchain decisions with real context from your codebase.

Language Framework

Priority Tradeoffs

Bring together product, content, and research agents to rank competing features against deadlines and resource constraints.

Roadmap Capacity

Code Review Debates

Stage a structured review where multiple coding agents critique a design proposal and vote on approval before a single line is written.

Review Consensus

Product Direction

Run a multi-agent strategy session to align on long-term goals, evaluate market positioning, and surface blind spots.

Strategy Alignment

Risk Assessment

Have security, research, and coding agents independently surface risks, then consolidate findings into a single prioritised report.

Security Due Diligence

Consensus Logic

Three Outcomes, Zero Ambiguity

Stance markers are parsed deterministically — no LLM interpretation, no hallucinated consensus.

FULL CONSENSUS

All agents AGREE

Every participating agent returns AGREE. The meeting exits immediately and minutes are written.

MAJORITY CONSENSUS

≥ 2/3 AGREE, 0 DISAGREE

A supermajority agrees with no opposition. The meeting exits early and records the majority verdict.

NO CONSENSUS

Neither condition met

Discussion continues to the next round. History is compressed and the coordinator re-frames the question.

Coming Soon

Join the Waitlist

Be the first to know when this plugin launches.

Join the Waitlist

Back to Plugins