multi

wt

Manage LlamaFarm worktrees for isolated parallel development. Create, start, stop, and clean up worktrees.

830 stars

Installation

Claude Code / Cursor / Codex

$curl -o ~/.claude/skills/wt/SKILL.md --create-dirs "https://raw.githubusercontent.com/llama-farm/llamafarm/main/.claude/skills/wt/SKILL.md"

Manual Installation

  1. Download SKILL.md from GitHub
  2. Place it in .claude/skills/wt/SKILL.md inside your project
  3. Restart your AI agent — it will auto-discover the skill

How wt Compares

Feature / AgentwtStandard Approach
Platform SupportmultiLimited / Varies
Context Awareness High Baseline
Installation ComplexityUnknownN/A

Frequently Asked Questions

What does this skill do?

Manage LlamaFarm worktrees for isolated parallel development. Create, start, stop, and clean up worktrees.

Which AI agents support this skill?

This skill is compatible with multi.

Where can I find the source code?

You can find the source code on GitHub using the link provided at the top of the page.

SKILL.md Source

# wt - Worktree Manager Skill

Manages isolated LlamaFarm development environments using git worktrees. Each worktree has its own services, ports, and data directories - enabling parallel agent sessions without conflicts.

## Full Documentation

For complete documentation, architecture details, and advanced usage:
@scripts/wt/README.md

---

## Quick Reference

| Task | Command |
|------|---------|
| Create worktree and start services | `wt create feat/my-feature` |
| Create and cd into worktree | `wt create --go feat/my-feature` |
| List all worktrees with status | `wt list` |
| Check service health | `wt status` or `wt health` |
| Start/stop services | `wt start` / `wt stop` |
| View service logs | `wt logs [server\|rag\|runtime\|designer\|all]` |
| Delete worktree | `wt delete <name>` |
| Switch to worktree | `wt switch <name>` |
| Open Designer in browser | `wt open` |
| Diagnose issues | `wt doctor` |
| Clean orphaned data | `wt gc` |

---

## When to Use wt

Use `wt` when:
- Starting isolated work on a feature/task that needs running services
- Running parallel coding sessions (multiple agents or terminals)
- Testing changes without affecting the main development environment
- Avoiding port conflicts between concurrent LlamaFarm instances

---

## Common Workflows

### Starting a New Task

```bash
# Create isolated environment with services running
wt create --go feat/my-task

# Work in the worktree...
# Services are already running on auto-assigned ports

# Check status anytime
wt status

# View logs if needed
wt logs server
```

### Checking What's Running

```bash
# List all worktrees with their port assignments
wt list

# Example output:
# NAME              STATUS    SERVER   DESIGNER  RUNTIME
# feat-my-task      running   8150     5150      11150
# fix-bug           stopped   8234     5234      11234
```

### Cleaning Up

```bash
# Stop services and remove a worktree
wt delete feat-my-task

# Remove data for worktrees that no longer exist
wt gc

# Remove worktrees for branches merged to main
wt prune
```

### Troubleshooting

```bash
# Diagnose common issues (ports, stale PIDs, missing tools)
wt doctor

# Restart stuck services
wt stop && wt start

# Force delete if normal delete fails
wt delete my-worktree --force
```

---

## Service URLs

Each worktree gets unique ports. Check URLs with:

```bash
wt url
# Outputs:
#   Server:   http://localhost:8150
#   Designer: http://localhost:5150
#   Runtime:  http://localhost:11150
```

If the Caddy proxy is running, use port-free URLs:
```
http://server.feat-my-task.localhost
http://designer.feat-my-task.localhost
```

---

## Key Environment Details

- **Worktrees location**: `~/worktrees/llamafarm/`
- **Data directories**: `~/.llamafarm/worktrees/<name>/`
- **Port allocation**: Deterministic hash of worktree name (14345+offset, 5000+offset, 11000+offset)
- **Logs**: `~/.llamafarm/worktrees/<name>/logs/`

---

## Notes for the Agent

1. **Always use `wt create --go`** when setting up a new task environment - it handles everything (branch, deps, build, services)
2. **Check `wt list` first** before creating a new worktree to see what already exists
3. **Use `wt status`** to verify services are healthy before running tests or making API calls
4. **Run `wt doctor`** when encountering unexplained service issues
5. **Clean up with `wt delete`** when a task is complete to free resources