tmux-dispatch
github.com/joeyspagnoli/tmux-dispatch- Category
- AI Agents
- Rank
- No. 1392Tools index
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Agent Building
- Interfaces
- Agent Skill / Plugin
- Builder
- joeyspagnoli
- GitHub
- 1 stars
- Date
About
A Claude Code skill that allows coding agents to spawn and manage multiple Claude Code sessions in tmux windows. Enables a driver session to create child agents, send them prompts, and coordinate their work for parallel processing or research workflows.
What it does
It makes long-lived Claude Code conversations addressable through tmux, then adds dependable turn completion signaling. Each worker occupies one pane. A Stop hook writes the final response into a sentinel file, allowing the controlling session to sleep until completion instead of guessing from terminal output or fixed delays.
Why it's ranked here
The design is small, understandable, and aimed at a genuine orchestration failure: detecting when an interactive agent has finished. Stable pane identifiers and atomic sentinel writes support the central promise. However, manual setup, hardcoded conventions, absent tests, and permission bypass by default make this a capable pattern for experienced operators rather than a polished automation product.
What's good
The completion mechanism uses Claude Code's own Stop event instead of parsing a changing terminal screen. Pane identifiers prevent collisions across concurrent workers. Literal input handling preserves quotes and multiline prompts. Atomic file replacement prevents partial reads, while the hook payload exposes the final assistant message directly. The guide also recommends files for large, diffable results.
Tradeoffs
Installation requires an executable hook and a manually inserted absolute path because the JSON configuration expands neither home shortcuts nor environment variables. Default launches grant workers full tool access without approval. Boot readiness and workspace trust still need handling. Several project, session, and binary locations are conventions that users must replace when their setup differs. No automated tests are shown.
How to use it well
Use it when you already operate Claude Code inside tmux and need persistent, multi-turn workers for research, planning, refinement, or parallel independent tasks. Capture pane identifiers when creating workers, arm a fresh sentinel before every prompt, and have large outputs written to known files. Skip it for simple one-shot jobs. Remote client access is not enabled by default.
Technical notes+
SKILL.md supplies the tmux command reference and spawning conventions. references/driving.md specifies the arm, send, background wait, and read cycle. hooks/stop-sentinel.sh reads Stop-hook JSON, derives a target from TMUX_PANE or session_id, writes beneath /tmp/claude-stop-sentinels, and uses a temporary file plus mv for atomic publication. hooks/driver-child-settings.json registers the command hook but contains REPLACE_WITH_ABSOLUTE_PATH_TO_HOOK, which README.md requires users to replace. The documented spawn command passes --dangerously-skip-permissions; driver mode also passes --settings and captures #{pane_id}. No test directory appears in the supplied tree.
Observed
- License
- No license is identified in the supplied repository tree or text.
- Primary language
- Bash provides the executable hook logic; Markdown contains most of the repository content.
- Installation
- Copy the repository folder into the Claude Code skills directory, make the hook executable, and replace the settings placeholder with an absolute path.
- Interface
- Claude Code skill operated through tmux shell commands and a Claude Code Stop hook.
- Configuration
- Hook registration uses JSON; project location, tmux session name, and Claude Code binary location are documented conventions.
- Repository structure
- The supplied tree contains a skill guide, README, driving reference, one shell hook, and one JSON settings file. No test directory is shown.
Read from README.md, SKILL.md, references/driving.md, hooks/stop-sentinel.sh, hooks/driver-child-settings.json.
What it can do
Spawn new Claude Code sessions in tmux windows
Command to create new session → New tmux window with active Claude Code session
Send prompts to child Claude sessions
Text prompt and target session identifier → Prompt delivered to specified Claude session
Coordinate parallel processing across multiple Claude sessions
Multiple tasks or prompts to distribute → Tasks distributed and executed across child sessions
Wait for and collect replies from child sessions
Session identifier to monitor → Response text from specified Claude session
Manage multiple tmux windows with Claude sessions
Session management commands → Active or terminated Claude Code sessions in tmux
Execute research workflows across distributed Claude agents
Research tasks and coordination instructions → Compiled research results from multiple agents
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.