dispatch

Solid

Orchestrate other AI coding agents (Claude Code, Codex, Cursor, OpenCode, Pi, Gemini, or any custom agent) by driving them through `ccmux invoke`. ccmux is the cross-harness substrate; YOU are the router. Use this skill when a prompt asks you to coordinate, delegate, fan out, or pipeline work across multiple agents, e.g. "plan with claude, implement with codex, search with gemini", "run these three agents in parallel and combine the results", "delegate this long implementation to codex while I keep working", "have another agent do X and summarize it back", or any request to use `ccmux invoke` to launch and watch worker agents. The user supplies the agent-per-task policy in their prompt; this skill teaches the mechanics of firing, polling, joining, cancelling, and reading worker output, plus when to hand a job off to `ccmux spawn` (a live, human-driven pane) instead of invoking it.

AI & Automation 104 stars 8 forks Updated today MIT

Install

View on GitHub

Quality Score: 86/100

Stars 20%
67
Recency 20%
100
Frontmatter 20%
70
Documentation 15%
100
Issue Health 10%
50
License 10%
100
Description 5%
100

Skill Content

# Orchestrating agents with `ccmux invoke` You are the orchestrator. ccmux gives you one uniform CLI that launches and observes every harness the same way (claude, codex, cursor, opencode, pi, gemini, plus custom agents). This skill teaches the **mechanics** of driving those workers. It is deliberately generic: the **agent-per-task policy comes from the user's prompt** ("plan with claude, implement with codex, search with gemini"), not from this file. Your job is to apply judgment to that policy and thread results between steps. ccmux never runs a model itself, not even to summarize a worker's output. Digesting a large result is your job (prompt the worker for brevity, or spawn a cheap summarizer over its full output). Keep that in mind throughout: the discipline that keeps an orchestration tractable is controlling how much each worker hands back. ## When to use - The user names a multi-agent workflow ("plan with X, implement with Y, search with Z"). - You need to fan a task out across several agents and combine their answers. - You want to delegate a long task to another agent without blocking your own work. - You want a second agent's independent take (review, verify, alternative implementation). ## When NOT to use - A single quick turn you can run inline: just `ccmux invoke <agent> "..."` once; you don't need the fire-and-poll machinery below. - Work you should do yourself. Delegation costs a cold start (5-15s) plus the worker's own round-trip; don't farm out wha...

Details

Author
epilande
Repository
epilande/ccmux
Created
3 weeks ago
Last Updated
today
Language
TypeScript
License
MIT

Bundled in these plugins

Similar Skills

Semantically similar based on skill content — not just same category

Web & Frontend Listed

orchestmux

Run several coding-agent CLIs (codex, kimi, claude, opencode, gemini) in parallel as tmux panes and coordinate them with the `orchestmux` CLI: create tasks, dispatch them to workers, block on real completion, and answer blocking questions. Use when the user asks to run agents in parallel, delegate work to codex/kimi/another agent while supervising it, split a job across workers, watch agents work in tmux, or says "orchestmux". Do not use for a single one-off shell command, or for work the current agent should just do itself.

4 Updated 3 days ago
layl-labs
AI & Automation Featured

dmux-workflows

Multi-agent orchestration using dmux (tmux pane manager for AI agents). Patterns for parallel agent workflows across Claude Code, Codex, OpenCode, and other harnesses. Use when running multiple agent sessions in parallel or coordinating multi-agent development workflows.

233,821 Updated today
affaan-m
AI & Automation Listed

orchestrate

Tech-lead orchestrator mode. Plan a goal, decompose it into subtasks, and automatically hand each subtask to the best CLI coding model you actually have installed — a strong reasoner for hard thinking, a fast worker for boilerplate/tests/bulk edits, and an independent model from a different vendor for a fresh perspective — then verify and synthesize. Detects available CLIs (codex, gemini, opencode, claude, and more) at runtime and always falls back gracefully, so it never hard-fails. Use when asked to "orchestrate", "be the tech lead", "coordinate this", "plan and delegate", or when a goal is big enough that one model doing everything sequentially would be slow or low-quality.

0 Updated 2 days ago
Vintaix