tmux-agents_skill

Overview

This skill manages background coding agents inside tmux sessions so you can run long-running coding tasks without blocking your terminal. It supports cloud agents (Claude, Codex, Gemini) that use API credits and local agents via Ollama that run on your machine for free. Use simple scripts to spawn, monitor, attach to, send instructions to, and kill agent sessions.

How this skill works

The skill launches a tmux session per agent using spawn.sh with a name, task, and agent choice. status.sh lists running sessions and check.sh inspects a session’s output; tmux attach lets you watch live. Local agents run models via Ollama on your Mac (uses GPU, no API costs); cloud agents use external APIs and credits.

When to use it

• Long-running refactors or project-wide changes you want to run in the background
• Parallel workstreams where multiple agents can operate simultaneously
• Experimentation and risky tasks where local models reduce cost
• Production-critical reviews or complex coding tasks requiring cloud models
• Quick edits or auto-approve tasks when you need fast results

Best practices

• Name sessions clearly (e.g., backend-fix, docs-write) to keep track of work
• Prefer local Ollama agents for expensive, experimental, or iterative tasks
• Use cloud agents for production-sensitive reviews or when you need higher-quality outputs
• Check status regularly with status.sh and attach only when you need to observe or intervene
• Kill sessions when finished to free system resources

Example use cases

• Spawn a Claude Code session to refactor the auth module overnight and review results in the morning
• Run an Ollama-based codex agent to iterate on unit tests locally without API costs
• Launch multiple agents in parallel for frontend, backend, docs, and tests to speed delivery
• Attach to a session to watch a complex migration run live and send follow-up instructions
• Use a cloud Gemini session to generate research and documentation while local agents handle code edits