harness-party-mode

To run multi-agent discussions for architecture reviews, orchestrate independent roundtable agents using your repository context. This coordinates multiple expert perspectives to analyze design questions and produce distinct, non-performative viewpoints directly from your project configuration.

Roundtable multi-agent orchestration spawns independent subagents to simulate a panel of experts discussing a complex engineering or product question. You need it when a single AI perspective is insufficient and you require genuinely distinct viewpoints, including disagreement, for cross-agent analysis.

Configure the agent roster by merging configured layers via resolve_config.py, which presents the available lineup before discussion. Load project context by injecting an optional project-context.md file, ensuring the multi-perspective analysis is aware of your specific repository details.

Yes, you can use multi-agent collaborative reasoning without subagents by adding the --solo flag. This enables solo mode, where the system roleplays each agent's full unabridged response in its own persona rather than spawning parallel subagents.

Yes, multi-agent roundtable discussion supports targeted multi-voice workflows including cross-talk and follow-ups. You can ask one agent to react to another agent's response, plus rotate agents across multiple discussion rounds for dynamic collaborative reasoning.