Headless multi-stage code change pipeline for Claude Code. Research, plan, implement, and merge autonomously via tmux and git worktrees.
# Add the CodefiLabs marketplace (one-time)
/plugin marketplace add codefilabs/marketplace
# Install autorun
/plugin install autorun@codefilabsAutorun requires tmux, python3, git, and the claude CLI.
Check your setup:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/doctor.shInstall missing dependencies:
bash ${CLAUDE_PLUGIN_ROOT}/scripts/doctor.sh --installOr install tmux directly:
# macOS
brew install tmux
# Ubuntu/Debian
sudo apt-get install tmux| Command | Description |
|---|---|
/autorun:start <task> |
Smart entry point — triages by scope (QUICK/MEDIUM/LARGE/EPIC) and routes to the right pipeline |
/autorun:research_codebase <task> |
Parallel codebase research with specialized sub-agents |
/autorun:create_plan <task or handoff> |
Create a phased implementation plan through iterative research |
/autorun:implement_plan <plan-path> |
Execute a plan phase-by-phase with verification teams |
/autorun:orchestrate <master-plan> |
Parse a staged master plan and launch parallel stages via git worktrees |
/autorun:merge <orch-dir> <stage> |
Merge a completed stage branch, launch next wave if ready |
/autorun:start "add caching to the API layer"
|
[triage: QUICK / MEDIUM / LARGE / EPIC]
|
LARGE → research_codebase → create_plan → implement_plan
MEDIUM → create_plan → implement_plan
QUICK → implement_plan (inline plan)
EPIC → interactive brainstorm → orchestrate (parallel stages)
Each pipeline stage chains to the next in a fresh tmux window, giving every phase a clean context window. For EPIC tasks, orchestrate launches multiple stages in parallel git worktrees — each running its own research/plan/implement pipeline — and merge handles branch integration and wave progression.
All runtime state lives in ~/.autorun/:
~/.autorun/
research/ # Research documents
plans/ # Implementation plans
review/ # Human-in-the-loop review files
orchestration/ # Per-plan stage status and context
logs/ # tmux session logs
MIT