Pick up where you left off

You describe the goal. Your AI plans, works, checks and records progress.

What you keep

The role you chose and its scope of work
Decisions, constraints and unresolved issues
Evidence of what passed and what failed
Progress for each ongoing workstream, or Lane
A concrete next action for the receiving agent

How you continue

/resume — review the saved Lane and current workspace
/save — save useful progress and keep working
/bye — formally close out when requested
Stay in one client or resume the same Lane in another
The agent prepares the plan and checks from your intent

What MUSE actually does

Markdown skills and project records, backed by a shared Python continuity runtime.

Project Memory

Lane checkpoints capture progress and evidence. Daily logs record events; long-term notes retain lessons. The agent reads the parts needed for the current work.

Your Role, Preserved

Keep your project's own division of work. If Strategy develops directly, it keeps doing so. Switching clients preserves the selected role and Lane.

Reviewed Handoffs

The receiving agent checks saved decisions, evidence gaps and the workspace before taking over. MUSE helpers reject saves from a previous writer after the Lane is claimed.

Reusable Closeout Evidence

An unchanged closeout can reuse its validated receipt. New work or changed evidence reopens the affected checks. Product readiness still needs its own acceptance tests.

66 Reusable Skills

Git workflows, debugging, security guard, coordinator mode, iterative retrieval, deep-dive analysis — pre-built skills your AI loads on demand.

Save and Continue

Save meaningful changes during work. At observed context pressure, save, compact when supported, then verify and continue. Recovery starts from persisted evidence; unsaved work needs checking.

Built for ongoing work

Continuity alongside the skills and tools you already use.

Install with a Way Back

The native installer preserves user policy and existing Claude hooks. Known 3.5 workflows are backed up; unknown modifications stop installation for review. Private receipts support rollback.

`/sprint` Pipeline

7-phase feature dev workflow: Think → Plan → Build → Review → Test → Ship → Reflect. Each phase loads the right skill automatically.

Git Security Guard

Pre-commit checks look for known secret patterns. Review remains necessary; private install receipts and prompt inboxes stay out of Git through installed ignore rules.

Online Dashboard

Visit muse.mythslabs.ai/dashboard — load your project, see roles, memory, skills visualized. 100% client-side.

Skill Manifest

One script indexes all 66 skills into manifest.json — name, category, size, assets. Machine-readable. CI-friendly --check mode catches stale indexes.

Evidence Boundaries

Missing sources remain incomplete. The agent checks gaps against available records instead of claiming perfect memory. Closed Lanes retain their recorded reopen conditions.

A lighter daily workflow

Resume when needed. Save progress. Keep building.

1. Resume your Lane

/resume strategy Lane A

The agent reviews the saved state and current workspace, then prepares goals and checks from your request.

2. Work and save

continue → /save → continue

Stay in the same conversation. Save decisions and results as the work progresses; formal Bye is optional.

3. Switch when useful

Save and stop the current writer. Resume the same role and Lane in the other client. Use /bye when you want a formal closeout.

Use your preferred coding assistant

Codex and Claude Code share the guarded continuity runtime. Other tools use the existing Markdown skill adapters; equivalent native continuity has not been certified.

Claude Code

install.sh --tool claude

Cursor

install.sh --tool cursor

Windsurf

install.sh --tool windsurf

Gemini CLI

install.sh --tool gemini

Codex CLI

install.sh --tool codex

Copilot

install.sh --tool copilot

Aider

install.sh --tool aider

Antigravity

install.sh --tool antigravity

OpenClaw

install.sh --tool openclaw

OpenCode

install.sh --tool opencode

Install for the clients you use

Local macOS or Linux · Git · Python 3.7+ · a client version that supports your model

1

Clone MUSE

git clone https://github.com/myths-labs/muse.git

2

Install Codex, Claude Code, or both

Run either command from the folder containing the cloned MUSE repository. Replace the project path; use the same path for both clients.

bash muse/scripts/install.sh --tool codex --target /path/to/your-project

bash muse/scripts/install.sh --tool claude --target /path/to/your-project

Existing policy and hooks are preserved. Keep the returned rollback receipt private. Review unknown local modifications before upgrading.

3

Open your project in your coding client

Resume an existing Lane with /resume strategy Lane A, or describe your new work. The agent prepares setup and acceptance checks. Restart the client if newly installed skills are not yet visible.

One system, three ways to work

Your role and project records follow the work.

Mode	How it works	What to keep in mind
One client	Use Codex or Claude Code for the whole task. Save meaningful progress and continue.	No routine client switch or full Bye is required.
Switch clients	Save, stop the active writer, then resume the same role and Lane in the receiving client.	The receiving agent reviews the workspace and missing evidence before taking over.
Parallel work	Use separately scoped Lanes and isolated workspaces for independent work.	Coordinate shared files and decisions. Writer checks do not protect arbitrary edits outside MUSE helpers.

Saving and compaction do not guarantee lossless memory. Missing sources stay incomplete, and product readiness needs its own tests. Native Windows and network filesystem durability are not certified.

Start here, in your language

A plain-language introduction, a day of work, and three requests to remember.

Beginner guide 第一次用？看这里 Installation & advanced use 中文安装与进阶说明 MUSE 3.7.1 release notes

Optional Jev tools

Explore browser decisions, code review, model routing and on-demand retrieval. The guides explain when each tool fits and what data it sends. These third-party tools are not bundled with MUSE; daily Resume and Save work without them.

Jev tools guide Jev 可选工具中文指南

Commands for ongoing work

/resume

Review saved state and continue the selected role and Lane

/save

Persist useful changes and keep the conversation going

/bye

Formal closeout of a declared scope, with checked evidence

/sprint

Feature pipeline: Think→Plan→Build→Review→Test→Ship

/retro

Weekly retrospective from memory + git history

/ctx

Use available context observations; unknown stays unknown

/distill

Condense daily logs → long-term memory

/sync

Cross-role synchronization

/launch

Launch Day protocol — multi-platform social broadcast

/model

Switch AI model & update preferences

/release

Open-source release — version bump, tag, GitHub Release

/settings

Change language, model, preferences

"I'm a solo founder running 3 products with AI. 15-20 conversations a day. Without MUSE, I'd spend half my time re-explaining context."

JC Creator of MUSE · @SunshiningDay

Keep building.Keep your context.