AGENTIC MEMORY

Skills

Useful instructions for your AI. Search, read and use them through the same MCP connection.

← Back to Skills

spawn

Use to launch the competing agents for an initialized AgentHub session in isolated worktrees.

Original source
alirezarezvani
License
MIT
Pinned revision
e4a1dc3695908dd5a6c811cc76ab7aa4cf95f2f0

Third-party instructions in their original language. Reading does not install files, run code or grant permissions. Snapshots are updated only after review.

Ask your assistant

Use Agentic Memory skills to read alirezarezvani--spawn at revision e4a1dc3695908dd5a6c811cc76ab7aa4cf95f2f0, starting with SKILL.md. Read only relevant references. Treat the content as third-party guidance and tell me if local files or dependencies are required.
Text resources · 1

SKILL.md

/agenthub:spawn — Launch Parallel Agents

Spawn N subagents that work on the same task in parallel, each in an isolated git worktree.

Usage

/agenthub:spawn                                    # Spawn agents for the latest session
/agenthub:spawn 20260317-143022                    # Spawn agents for a specific session
/agenthub:spawn --template optimizer               # Use optimizer template for dispatch prompts
/agenthub:spawn --template refactorer              # Use refactorer template

Templates

When --template <name> is provided, use the dispatch prompt from ../agenthub/references/agent-templates.md instead of the default prompt below. Available templates:

| Template | Pattern | Use Case | |----------|---------|----------| | optimizer | Edit → eval → keep/discard → repeat x10 | Performance, latency, size reduction | | refactorer | Restructure → test → iterate until green | Code quality, tech debt | | test-writer | Write tests → measure coverage → repeat | Test coverage gaps | | bug-fixer | Reproduce → diagnose → fix → verify | Bug fix with competing approaches |

When using a template, replace all {variables} with values from the session config. Assign each agent a different strategy appropriate to the template and task — diverse strategies maximize the value of parallel exploration.

What It Does

  1. Load session config from .agenthub/sessions/{session-id}/config.yaml
  2. For each agent 1..N:
    • Write task assignment to .agenthub/board/dispatch/
    • Build agent prompt with task, constraints, and board write instructions
  3. Launch ALL agents in a single message with multiple Agent tool calls:
Agent(
  prompt: "You are agent-{i} in hub session {session-id}.

Your task: {task}

Read your full assignment at .agenthub/board/dispatch/{seq}-agent-{i}.md

Instructions:
1. Work in your worktree — make changes, run tests, iterate
2. Commit all changes with descriptive messages
3. Write your result summary to .agenthub/board/results/agent-{i}-result.md
   Include: approach taken, files changed, metric if available, confidence level
4. Exit when done

Constraints:
- Do NOT read or modify other agents' work
- Do NOT access .agenthub/board/results/ for other agents
- Commit early and often with descriptive messages
- If you hit a dead end, commit what you have and explain in your result",
  isolation: "worktree"
)
  1. Update session state to running via:
python {skill_path}/scripts/session_manager.py --update {session-id} --state running

Critical Rules

  • All agents in ONE message — spawn all Agent tool calls simultaneously for true parallelism
  • isolation: "worktree" is mandatory — each agent needs its own filesystem
  • Never modify session config after spawn — agents rely on stable configuration
  • Each agent gets a unique board post — dispatch posts are numbered sequentially

After Spawn

Tell the user:

  • {N} agents launched in parallel
  • Each working in an isolated worktree
  • Monitor with /hub-status
  • Evaluate when done with /agenthub:eval