fork-terminal

name: fork-terminal
description: Use when distributing independent tasks across multiple AI coding assistants in parallel. Triggers include "fork this work", "spawn agents", "run in parallel", "distribute tasks", or when multiple independent tasks would benefit from concurrent execution.

enable_claude_code: true
enable_gemini_cli: true
enable_aider: true
default_agent: claude
session_tracking: true
headless_mode: false

How Fork-Terminal Works

This skill spawns independent terminal sessions with structured context handoffs. Each spawned session receives a handoff document containing task, context, and constraints.

1. Handoff Documents Are Critical

Never spawn a terminal without a proper handoff document. Raw commands work for CLI tools, but agentic tools need structured context:

• What task to complete
• What files/code to focus on
• What constraints apply
• What success looks like

2. Track What You Spawn

Every spawned session gets logged to a manifest file. This enables:

• Status checks across all spawned work
• Resumption if sessions fail
• Coordination between parallel tasks

3. Match Tool to Task

4. Interactive vs Headless

• Interactive: User takes over the terminal (default for complex tasks)
• Headless: Agent runs autonomously, output captured (for batch work)

Set headless_mode: true in variables for autonomous execution.

What would you like to do?

1. Fork work to agents - Distribute tasks across Claude Code, Gemini, or Aider
2. Run CLI command - Spawn a raw terminal command
3. Check status - See status of spawned sessions
4. Custom dispatch - Specify exactly what to spawn and how

Wait for response before proceeding.

| Response | Workflow | Notes |
|----------|----------|-------|
| 1, "fork", "spawn agents", "distribute", "parallelize" | workflows/analyze-and-dispatch.md | Main workflow - analyzes task, creates handoffs |
| 2, "CLI", "command", "terminal", "run" | workflows/spawn-cli-command.md | Simple command dispatch |
| 3, "status", "check", "progress", "sessions" | workflows/check-status.md | Review spawned work |
| 4, "custom", "specific" | workflows/spawn-agent.md | Direct agent dispatch |

Variable-based routing:

• If user requests Claude Code AND enable_claude_code is false → Suggest alternative
• If user requests Gemini AND enable_gemini_cli is false → Suggest alternative
• If user requests Aider AND enable_aider is false → Suggest alternative

After reading the workflow, follow it exactly.

Domain Knowledge

All in references/:

Agent Patterns: agent-invocations.md (how to invoke each tool)
Handoff Design: context-handoff-patterns.md (structuring handoffs)
Session Management: session-tracking.md (tracking spawned work)

Workflows

Meta-Prompts

Templates

Quick Start

Fork work to Claude Code:

User: "Fork this to Claude Code - implement the login form in src/components/"

Parallel dispatch:

User: "Spawn 3 agents - one for frontend, one for API, one for tests"

Check progress:

User: "Check status of spawned sessions"

A successful fork-terminal invocation:

• Creates structured handoff document for each spawned agent
• Logs session to manifest for tracking
• Launches appropriate tool (CLI or agent) in new terminal
• Returns confirmation with session ID and how to check status