Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Runs multiple Cursor agents in parallel using tmux sessions, splitting large coding tasks into non-conflicting subtasks handled simultaneously. A second mode, called council, assigns different AI models (Claude Opus, Sonnet, GPT) to roleplay named technical personas and analyze a decision from competing viewpoints, then synthesizes the outputs.
Parallel tmux sessions cut wall-clock time for multi-module tasks, and assigning distinct model personas to the council produces genuine perspective diversity rather than one model self-agreeing across prompts.
When to use it
- Refactoring frontend components while adding backend error handling at the same time
- Getting architecture opinions from simulated competing engineering philosophies
- Running test coverage work in parallel with new feature implementation
- Evaluating a database or queue selection through structured multi-model debate
- Coordinating independent feature branches without waiting for sequential agent completion
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Which message queue should we adopt — Redis Streams, RabbitMQ, or NATS — for a 500k messages/day system run by a 3-person team?
- 1Write the problem context and constraints to a shared markdown file
- 2Create three tmux sessions and assign each a persona prompt: Opus as Joe Armstrong (fault tolerance), Sonnet as TJ Holowaychuk (simplicity), GPT as Ryan Dahl (design regret)
- 3Send each prompt to its session using tmux send-keys and pipe output to files
- 4Poll sessions until all three finish, then collect the output files
- 5Synthesize the three perspectives into a final recommendation and archive to ~/.openclaw/workspace/pr-review/
Three distinct expert analyses with a synthesized recommendation, saved as persistent session files