Let Claude Code and Codex run your dev environment.

# Every project, its running state, and active agents
lpm list --json

# One project's services, actions, and live status
lpm project myapp

# Live agent status across projects
lpm status

# Start the whole project, or just one profile
lpm start myapp
lpm start myapp --profile backend
lpm stop myapp

# Restart a single dev server
lpm service api restart

# Last 200 lines of a running service
lpm logs frontend -n 200

# Block until a port answers — no blind sleep loops
lpm wait --port 3000 --timeout 60

# Open Claude Code in a new lpm terminal with a prompt
lpm run claude --prompt "fix the failing spec"

# Flag the project as waiting on you in the sidebar
lpm set-status review Waiting

exit 0 success·exit 2 not found / app not running·exit 1 error / timeout

# Clone into 3 copies and run the same prompt in each
lpm duplicate -n 3 --run claude \
  --prompt "make the checkout flow pass its tests"

# Block until a copy's agent settles (names come from the output above)
lpm wait myapp-k7q2xm --agent

# Keep the winner, remove the rest
lpm remove myapp-3fz8ta

1. 01

   Fan out

   lpm duplicate -n 3 makes three real, standalone copies of the project (a fast copy-on-write clone on Apple’s filesystem), lists them under the original in the sidebar, and queues the same agent and prompt in each. Prefer branches? lpm worktree takes the same flags and makes linked Git worktrees instead.

2. 02

   Let them race

   Three agents attack the same task in parallel, each in its own copy with its own files, terminals, and services. lpm wait <copy> --agent blocks on each copy until its agent has finished, with no polling loops.

3. 03

   Keep the best, clean up

   Compare the results, keep the copy you like, and lpm remove tidies up the rest. Copies are ordinary folders, with nothing to unwind and no shared state to untangle.

# Auth refactor

## Goal
Move sessions from cookies to signed tokens.

## Current state
Token issuing works; refresh flow is next.
Blocked on the mobile client's token storage.

## Timeline
### 2026-09-22 09:40 — claude
- Done: issue and verify tokens in the API
- Next: refresh endpoint

### 2026-09-23 14:05 — codex
- Done: refresh endpoint and tests
- Open: where the mobile app keeps the token

Memory is read only when you ask for it, so nothing is added to every session on its own. Sessions stay on your computer, separate from each CLI's built-in memory, and are available for local projects.

• Can Claude Code restart my dev server?

  Yes. Once the lpm skill and CLI are installed, Claude Code can run lpm service <name> restart to bounce a single dev server, or lpm start / lpm stop for the whole project — then lpm wait to block until it is ready again.
• Which AI coding agents does this work with?

  Claude Code, Codex, Gemini CLI, and OpenCode. The skill is installed both for Claude Code and for the open agent-skills directory that Codex, Gemini CLI, and OpenCode read, so the same lpm commands work across all of them.
• Do I need MCP to connect agents to lpm?

  No. Agents drive lpm through a normal command-line tool. One click in Settings installs the skill and the lpm CLI together — there is no server to run and no MCP configuration.
• How does an agent know it is inside an lpm project?

  Every terminal inside lpm is tagged with the project it belongs to. Ask the agent to restart a server, read logs, or fan out, and it reaches for the lpm skill; the tag then lets it leave out the project name. No per-project setup is needed.
• Can I invoke the skill manually?

  Yes. Type /lpm in Claude Code, or ask any other agent to use the lpm skill, to load both lpm skills at once: project control and config editing. It's useful in a terminal outside lpm, or whenever you want the skill in explicitly instead of waiting for the agent to pick it up.
• Can I run multiple AI agents in parallel on copies of a project?

  Yes. lpm duplicate clones a project into real standalone copies (up to 50) and can queue the same agent and prompt in each; lpm worktree does the same with linked Git worktrees. lpm wait <copy> --agent blocks on a copy until its agent finishes, and lpm remove cleans up the copies you do not keep.
• Should an agent call lpm duplicate or lpm worktree?

  Reach for lpm duplicate if the agent must see your working state: node_modules, .env files and uncommitted edits come along, and flags can drop the edits or reinstall dependencies. Use lpm worktree when committed code is enough: it branches off your current commit, costs little disk and merges back easily, but ignored and uncommitted files are missing.
• Can Claude Code and Codex share context?

  Yes, through the lpm-memory skill. An agent saves a named work session with its goal, current state, and a timeline, and any other agent can continue it by name, for example /lpm-memory auth-refactor. Memory is read only when you ask for it and lives on your computer, per project.
• Can I manage automations from the command line?

  Yes. lpm automations lists your scheduled agent jobs, runs or stops one, pauses or resumes its schedule, prints its history or live output, and replies to an AI automation's conversation. Creating and editing jobs happens in the app, and the bundled skills don't cover automations, so tell an agent about these commands if you want it to use them.
• Can agents read my dev-server logs?

  Yes. lpm logs <service> returns the recent output of any running service, so an agent can see the error it just caused and fix it. Nearly every command also supports --json for structured output, and exit codes are agent-friendly.
• Does this run on Windows or Linux?

  Yes. Besides macOS, the desktop app is in beta for Windows 11 (x64) and for 64-bit x86 Linux (.deb, .rpm and AppImage). On Windows, install Git for Windows first, since lpm runs terminals, services and agents through Git Bash. The lpm command-line tool and the agent skills work the same on all three. A Linux server can also be an lpm host: the same lpm runs headless on an x86_64 server with Ubuntu 22.04 or newer (glibc 2.35+), and you add it from the desktop app by typing user@host in Settings → Connections. That server's projects, terminals and agents then appear in your sidebar, and the lpm command-line tool is installed there too.