Party Mode Meeting#

Party Mode assembles a small group of available agents and facilitates a sequential roundtable. It does not execute tools, write files, call external providers, manage memory, or run autonomous workflows.

Command#

When the user invokes:

text

/party <topic>

start a Party Mode meeting for the topic. If /party has no topic, ask for the meeting topic before assembling agents.

Meeting Flow#

1. Read the meeting topic from the /party command.
2. Identify available agents from the OpenClaw agent context.
3. Select a focused group of 3-4 relevant agents.
4. List the selected agents before the meeting starts.
5. Invite agents one at a time in a deliberate order.
6. Preserve a short discussion context after every turn.
7. Allow each agent to respond, react, or pass.
8. Present the full meeting transcript to the user.
9. End with a brief convergence, divergence, and open-questions summary.

Group Assembly#

Choose agents based on the topic:

• Use the available OpenClaw agents already visible to the host environment.
• Do not invent agents that are not available.
• Lead with the agent whose expertise best anchors the topic.
• Add agents from different model families or reasoning styles when available.
• Prefer variety over duplication.
• If the user names a specific agent, include that agent.
• If fewer than three relevant agents are available, continue with the available agents.
• If more than four agents are relevant, choose the strongest four and mention that the group was kept small for coherence.

Before the first turn, announce:

text

I've brought together [agent list] to discuss: [topic].

Agent Turn Prompt#

For each agent, use this structure:

text

You are {name} ({title}), participating in a roundtable meeting.

Meeting topic:
{topic}

Discussion so far:
{short summary under 300 words, omitted for the first agent}

Guidelines:
- You may respond or pass.
- Start with: {name}:
- Do not repeat prior points unless you are adding a new angle.
- Build on, challenge, or clarify what came before.
- If the topic is outside your expertise, pass briefly.
- Do not use tools.

After each turn, update the discussion-so-far summary in plain language. Keep it under 300 words and include only the points needed by the next agent.

Reactions#

After each substantial response, ask the remaining agents whether they want to react. A reaction should be short and should either build on the point, challenge it, or pass.

Do not force a reaction from every agent. If an agent has nothing new to add, record that they passed and continue.

Final Output#

Return the meeting in this shape:

text

## Meeting: [topic]

Participants:
- [agent]
- [agent]
- [agent]

[Agent A response]

[Agent B response or pass]

[Optional short reactions]

[Agent C response or pass]

## Wrap-Up
[2-3 sentence summary of convergence, divergence, and open questions]