director

Coordinate parallel Claude Code worker sessions across tmux/rmux panes.

Updated Jun 4, 2026
One-click install
npx skills add https://github.com/rp4ri/orchestration-studio --skill director-rp4ri
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: director
Source: https://github.com/rp4ri/orchestration-studio/tree/main/skills/director
Command: npx skills add https://github.com/rp4ri/orchestration-studio --skill director-rp4ri

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Manually coordinating multiple parallel Claude Code worker sessions across tmux/rmux panes is error-prone, often leading to lost worker context, conflicting uncommitted file changes, and missed task completion signals that waste hours of development time.

Core Features & Use Cases

  • Worktree-per-worker isolation: Automatically creates separate git worktrees for each worker to prevent uncommitted file conflicts between parallel sessions.
  • Safe prompt dispatch protocol: Enforces an exact keystroke sequence to avoid ghost text submission and lost mandates that leave workers idle while tasks are stranded.
  • Background watcher system: Uses debounced state detection to reliably monitor worker completion without blocking the director's reactive workflow.
  • Context-preserving re-tasking: Allows redirecting workers without restarting their sessions to retain built-up project context and plans.
  • Use Case: Orchestrate 3 parallel workers to audit separate monorepo packages, with the director automatically merging their PRs in dependency order once all tasks complete.

Quick Start

Ask the director skill to set up a tmux session with 3 isolated worker panes, dispatch a separate code audit task to each worker in their own git worktree, and automatically merge their resulting PRs once all workers finish their work.

Frequently Asked Questions about director

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run parallel Claude Code workers in tmux without losing task context?

Parallel Claude Code orchestration uses debounced state detection and background watchers to reliably monitor worker completion across tmux panes without blocking the director's reactive workflow. This prevents missed task completion signals that waste hours of development time.

How do I prevent uncommitted file conflicts between parallel Claude Code sessions?

To prevent uncommitted file conflicts between parallel Claude Code sessions, implement worktree-per-worker isolation that automatically creates separate git worktrees for each worker pane. This ensures isolated worker environments for parallel code audits and monorepo maintenance.

Can I redirect a Claude Code worker to a new task without restarting its tmux session?

Yes, you can redirect a Claude Code worker without restarting its tmux session by using context-preserving re-tasking. This retains the worker's built-up project context and plans while allowing the director to dispatch new mandates across panes safely.

What's the best way to coordinate cross-platform releases across multiple Claude Code agents?

The best way to coordinate cross-platform releases across multiple Claude Code agents is applying a multi-agent orchestration layer with worktree-per-worker isolation and background watcher-based completion monitoring. This safely dispatches prompts and automatically merges resulting PRs in dependency order once all workers finish.

Does multi-agent orchestration for bulk PR triage require git worktrees for each tmux pane?

Yes, multi-agent orchestration for bulk PR triage requires git worktrees for each tmux pane to maintain worktree-per-worker isolation. This prevents conflicting uncommitted file changes and ensures safe prompt dispatch across isolated worker environments.

Why do parallel Claude Code workers in tmux sometimes miss task completion signals?

Parallel Claude Code workers in tmux miss task completion signals due to a lack of debounced state detection and background watcher-based completion monitoring. Without these mechanisms, the director's reactive workflow is blocked and completion signals are missed.