Orchestrates parallel work by dispatching tasks to AI sub-agents in isolated git workspaces.

Install

mkdir -p .claude/skills/subtask && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/6761" && unzip -o skill.zip -d .claude/skills/subtask && rm skill.zip

Installs to .claude/skills/subtask

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Parallel task orchestration CLI that dispatches work to AI workers (via Claude Code) in isolated git workspaces. Use when the user wants to draft, create, run, or manage tasks, delegate tasks to workers/subagents, or mentions subtask or Subtask.
245 chars✓ has a “when” trigger
Advanced

Key capabilities

  • Orchestrate parallel AI workers in isolated git worktrees
  • Draft, create, and manage tasks for sub-agents
  • Review and iterate on worker output
  • Merge task branches into base branches
  • Trace and debug worker internal thinking

How it works

It uses a CLI to dispatch work to AI workers running in isolated git worktrees, managing the workflow from drafting to merging.

Inputs & outputs

You give it
Task requirements and delegation instructions
You get back
Merged code changes from sub-agents

When to use subtask

  • Delegate research to a sub-agent
  • Manage parallel feature development
  • Draft and execute complex tasks across workers

About this skill

Subtask

Subtask is a CLI for orchestrating parallel AI workers. There are three roles: the user who gives direction, you (the lead) who orchestrates and delegates, and workers who execute tasks.

Each worker runs in an isolated git worktree. They can't conflict with each other.

The user tells you what they need. You clarify requirements, break work into tasks, dispatch to workers, review their output, iterate until it's right, and merge when ready.

Prefer to delegate exploration, research and planning to workers as parts of their tasks. Workers have time & space to dig deep, whereas you should preserve context to lead. Only go into details yourself when user explicitly requested, or the situation calls for it.

Mindset

  1. Understand before delegating — ask questions, clarify requirements. Don't rush to create tasks until you understand what the user actually wants.
  2. Own the complexity — stay on top of all tasks. Surface progress and blockers. Don't make the user chase status.
  3. Work autonomously — review output, request changes, iterate with workers. Only involve the user for decisions they need to make.
  4. Ask before merging — get user sign-off before merging. Don't merge without user approval.

Commands

CommandDescription
subtask ask "..."Quick question (no task, runs in cwd)
subtask draft <task> --base-branch <branch> --title "..." <<'EOF'Create a task
subtask send <task> <prompt>Prompt worker on task (blocks until reply)
subtask stage <task> <stage>Advance workflow stage
subtask listView all tasks
subtask show <task>View task details
subtask diff [--stat] <task>Show changes (from merge base)
subtask merge <task> -m "msg"Squash-merge task into base branch
subtask close <task>Close without merging, free workspace
subtask workspace <task>Get workspace path (a git worktree)
subtask interrupt <task>Gracefully stop a running worker
subtask log <task>Show task conversation and history
subtask trace <task>Debug what a worker is doing and thinking internally

Tip: Add --follow-up <task> on draft to carry forward conversation context from a prior task.

Flow

# 1. Draft (task name is branch name, task description is shared with worker)
subtask draft fix/bug --base-branch main --title "Fix worker pool panic" <<'EOF'
There's an intermittent panic in the worker pool under high concurrency—likely a race condition in pool.go.
Reproduce, find root cause, fix, and add tests.
EOF

# 2. Start the worker
subtask send fix/bug "Go ahead."

# 3. When worker finishes, review and iterate
subtask stage fix/bug review
# Review with `subtask diff --stat fix/bug`, or read the files at `cd $(subtask workspace fix/bug)`.

# 4. Request changes if needed
subtask send fix/bug <<'EOF'
Also handle the edge case when pool is empty.
EOF

# 5. When ready, merge or close
subtask stage fix/bug ready
subtask merge fix/bug -m "Fix race condition in worker pool"
# Or if not merging: subtask close fix/bug

Critical: Use the Bash tool with run_in_background: true for subtask send. Tell the user you're waiting and stop. Don't poll or check. You'll be notified when done.

Merging

subtask merge squashes all task commits into a single commit on the base branch.

subtask merge fix/bug -m "Fix race condition"

If conflicts occur, merge will fail with instructions. Follow them.

Stages

All tasks have stages: doing → review → ready

StageWhen to advance
doingWorker is working (default)
reviewWorker done, you're reviewing code
readyReady for human to decide (human review, merge, more work, etc.)

Advance with: subtask stage <task> <stage>

Planning Workflows

For complex tasks, add a plan stage: plan → implement → review → ready

You plan (--workflow you-plan): You draft PLAN.md in task folder, worker reviews and pokes holes. They plan (--workflow they-plan): Worker drafts PLAN.md in task folder, you review and approve or request changes.

Notes

  • Use subtask list to see what’s in flight.
  • Use subtask show <task> to see progress and details.
  • Use subtask log <task> to see task conversation and events.
  • Use subtask trace <task> to debug what a worker is doing and thinking internally.

When not to use it

  • For simple tasks that do not require parallel orchestration or isolated environments

Limitations

  • Requires user sign-off before merging
  • Requires manual conflict resolution if merge fails

How it compares

It enables parallel task execution in isolated environments rather than performing all work in a single workspace.

Compared to similar skills

subtask side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
subtask (this skill)36moReviewAdvanced
skillshare12moReviewIntermediate
monitor-tweet04moReviewAdvanced
agentic-jujutsu34moReviewAdvanced

Try saying

Example prompts that trigger this skill in your AI assistant.

You might also like

skillshare

runkids

Syncs skills across AI CLI tools (Claude, Cursor, Windsurf, etc.) from a single source of truth. Use when: "sync skills", "install skill", "search skills", "list skills", "show skill status", "backup skills", "restore skills", "update skills", "new skill", "collect skills", "push/pull skills", "add/remove target", "find a skill for X", "is there a skill that can...", "how do I do X with skills", "skillshare init", "skillshare upgrade", "skill not syncing", "diagnose skillshare", "doctor", or any skill/target management across AI tools.

116

monitor-tweet

ArthTech-AI

Run the tweet-to-toolkit pipeline on a tweet. Flow: (1) triage+feasibility, (2) ALWAYS show findings and BOTH repo options to user, (3) user approves target + direction, (4) build auto-starts, (5) review auto-starts, (6) if APPROVED, gh pr create auto-runs. The ONLY human gate is step 2 — everything

00

agentic-jujutsu

ruvnet

Quantum-resistant, self-learning version control for AI agents with ReasoningBank intelligence and multi-agent coordination

328

moai

modu-ai

MoAI super agent - unified orchestrator for autonomous development. Routes natural language or explicit subcommands (plan, run, sync, fix, loop, project, feedback) to specialized agents. Use for any development task from planning to deployment.

14

build-with-agent-team

coleam00

Build a project using Claude Code Agent Teams with tmux split panes. Takes a plan document path and optional team size. Use when you want multiple agents collaborating on a build.

13

swarm-coordination

joelhooks

Multi-agent coordination patterns for OpenCode swarm workflows. Use when work benefits from parallelization or coordination.

04

Search skills

Search the agent skills registry