Orchestration

Split independent tasks across parallel agents so their results merge cleanly

Decide when to run parallel sub-sessions for exploration or adversarial review

Coordinate multi-agent work with inboxes, threaded messages and file reservations

Coordinate parallel multi-agent teams that build large projects through git

List, message, wait on and stop Happier coding sessions from the command line

Check AgentPool capacity signals and run privacy-safe diagnostics before delegating

Launch and monitor Conductor Cloud agent sessions with bounded orchestration

Create and track asynchronous Jules coding tasks through its planning API

Manage Paperclip agent workflows through its REST API from the command line

Build and run durable JavaScript workflows that fan out work across Codex agents

Resume a coding session when a blocking GitHub issue gets its first new comment

Send and read messages between coding agent sessions in tmux panes

Communicate with other CLI agents and scripts in tmux panes

Mark a multi-step plan complete and produce its success and failure summary

Create a step-by-step plan before running any multi-step task with tools

Check which steps of the current plan are done, in progress or failed

Extend the timeout for long-running tasks so they are not cut off early

Mark a plan step as pending, in progress, completed, failed or skipped

Design autonomous Claude Code loops from simple pipelines to multi-agent DAGs

Create and use subagents and the Task tool to hand specialized work to agents

Set up the Ralph autonomous coding loop in a directory with prompts and backpressure

Run coding agents in an iterative ralph loop from a prompt or prompt file

Delegate coding tasks to Codex or Claude Code in background sessions with worktrees

Gather several independent perspectives on a question, then synthesize one answer

Review and revise a draft independently in a loop until it meets quality criteria

Improve existing agents through performance analysis, prompt tuning and iteration

Tune multi-agent systems for throughput, reliability and cost-aware workload splits

Write and run OpenProse programs that orchestrate multi-agent workflows in session

Create workflow commands that run multi-step tasks through sub-agents

Review work from several judge perspectives and build consensus on the verdict

Run a task through a sub-agent and retry it until an LLM judge approves the result

Have several agents compete on a task and synthesize the strongest result

Run independent tasks side by side with parallel sub-agents and judge checks

Run dependent steps in order, passing context forward and judging each step

Have a meta-judge and a judge agent evaluate the results from this conversation

Evaluate solutions with independent judges who debate in rounds until they agree

Launch a sub-agent with automatic model choice and a self-critique check

Design multi-agent architectures for tasks that exceed one agent context limit

Give each task to a fresh sub-agent and review code between tasks for quality

Run a simulated chatroom where expert personas debate a topic you choose

Simulate an Austrian economics debate between Hayek and Mises personas

Call the local Grok Build CLI from Codex and return its response

Guidance for handing focused work to child agents and subagents

Hand coding, review and refactor tasks off to the Claude Code CLI with approval

Delegate code implementation and reviews to OpenAI Codex models from your agent

Message between Claude Code, Codex and Gemini CLI agents through a local SQLite store

Send messages to OpenClaw, check its status and manage its config and skills

Read and control tmux panes so agents can message each other

Delegate coding tasks and code reviews to the local Codex CLI from your session