Launch Parallel Agents — Development

Launches N parallel subagents in isolated git worktrees to compete on the session task for an AgentHub session. It writes dispatch assignments, builds agent prompts, and spawns all agents in a single message for true parallelism.

engineering claude-code

📂 View skill on GitHub ↗ 🚀 How to use

✨ What it does

Loads session config from .agenthub/sessions/{session-id}/config.yaml.
Writes task assignments to .agenthub/board/dispatch/ for each agent.
Builds agent prompts with task, constraints, and board write instructions.
Launches all agents in a single message with multiple Agent tool calls, each in an isolated worktree.
Updates session state to 'running' via session_manager.py.
Supports --template flag to use dispatch prompts from agent-templates.md.

🎯 When to use it

When the user runs /hub:spawn to start competing agents for an initialized AgentHub session.
When you need to explore multiple approaches to a task in parallel.
When you want to use a specific template (optimizer, refactorer, test-writer, bug-fixer) to guide agent strategies.

🚀 How to use

Trigger: /hub:spawn

Trigger with the slash command /hub:spawn, optionally specifying a session ID and a template. The skill expects an initialized AgentHub session with a config file. Example prompts:

/hub:spawn
/hub:spawn 20260317-143022
/hub:spawn --template optimizer

📄 Output: Launches N parallel agents and updates session state; no files are produced directly, but dispatch posts are written to the board.

📦 Add this skill to Claude Code

# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills

# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering/agenthub/skills/spawn .claude/skills/spawn

Skill source: engineering/agenthub/skills/spawn/SKILL.md

⚠️ Good to know

Requires an initialized AgentHub session with a valid config.yaml; agents must not modify session config after spawn.

❓ FAQ

How are agents launched?

All agents are launched in a single message with multiple Agent tool calls, each with isolation set to 'worktree'.

What templates are available?

Available templates are optimizer, refactorer, test-writer, and bug-fixer, each with a specific pattern and use case.

What should agents do after finishing?

Each agent must write a result summary to .agenthub/board/results/agent-{i}-result.md including approach, files changed, metric if available, and confidence level.

🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.