muse

# MUSE — Continuous Development with Codex and Claude Code

> MUSE 3.7.1 preserves roles, decisions and next steps in shared project records, with complete-scope recovery for larger worktrees.

MUSE combines Markdown skills and workflows with a Python continuity runtime.
Use Codex alone, Claude Code alone, or move the same role and Lane between them.
A Lane is a workstream with its own checkpoint; the agent derives goals,
constraints and acceptance checks from the user's request and recorded evidence.

## Daily use

- Continue an active conversation with a normal request such as "continue".
- `/resume strategy Lane A`: review and recover the chosen role and Lane.
- `/save`: persist useful progress without ending the conversation.
- `/bye`: perform an explicit formal closeout for a declared scope.
- Context pressure: save, compact when supported, verify the state, then continue.
  A percentage does not force Bye or a new conversation. Unknown usage stays unknown.

Projects retain their own division of work. Strategy can implement directly;
starting technical work or changing clients does not automatically switch to Build.
Existing closed Lanes retain their reopen conditions. Missing old Lane records
must be investigated, not silently initialized as empty.

## Switching clients

Save and stop the active writer, then resume the same role and Lane in the other
client. The receiving agent reviews the workspace, decisions and evidence gaps
before taking over. MUSE helpers reject saves from a previous writer after a
claim. Parallel work needs distinct scopes and isolated workspaces; arbitrary
external file edits are outside these helper protections.

## Install or upgrade

Requires Git, Python 3.7+ and a local macOS/Linux filesystem. Use a client version
that supports the selected model. From the directory containing the cloned repo:

```bash
git clone https://github.com/myths-labs/muse.git
bash muse/scripts/install.sh --tool codex --target /path/to/project
bash muse/scripts/install.sh --tool claude --target /path/to/project
```

Install only the client you use, or point both at the same project. User policy
and existing Claude hooks are preserved. Known 3.5 workflows are backed up;
unknown local modifications stop installation for review. Installation does not
create or claim a Lane. Keep installation/rollback receipts private.

## Runtime environment checks

Version 3.7.1 adds an optional preflight for the Python interpreter a workflow
actually uses. It checks isolation and can verify and fully decode PNG/JPEG
fixtures with Pillow. Ordinary continuity remains standard-library-only; the
installer does not add Pillow. This is not a native attachment adapter or proof
that the real workflow completed. Rerun that workflow in the same environment.

## Checkpoint capacity

Incremental save input, selected source notes and checkpoints each have a 256 KiB
limit, measured as UTF-8 bytes including rendered metadata and separators. Git
full-scope batches do not expand it. Keep detailed evidence in referenced files
and use bounded notes and pointers; only Next Action and Required Reads may be
replaced. Preserve history, source coverage and writer identity. On TOO_LARGE,
retain the failed attempt, review a smaller supported delta and verify the guarded
save's archive and canonical readback. If it still cannot fit, retain unsaved
evidence for a separate storage review; do not raise limits or delete history.
This is guidance for existing behavior, not a new capacity feature or release.

## Optional Jev tools

Separate third-party projects can assist with browser decisions, code-review
triage or model routing in a separately launched CLI. They are not bundled MUSE
skills or an automatic integration. MUSE remains at 3.7.1 with 66 included skills;
its installer does not install these tools or enable provider calls. Ordinary
Resume and Save do not require Jev. The bilingual guides cover reviewed source
versions, outgoing data, unsupported tasks and an optional Lane retrieval pattern.
That pattern is not a shipped muse-context command. Installation checks and model
decisions do not certify real task outcomes.

## Package and compatibility

- 66 skills: Core 8, Toolkit 44, Ecosystem 13, Growth distribution 1.
- Guarded native continuity: Codex and Claude Code.
- Existing skill adapters: Claude Code, OpenClaw, OpenCode, Cursor, Windsurf,
  Gemini CLI, Codex, Copilot, Aider, Antigravity. Adapter support does not imply
  equivalent native continuity certification.
- Optional project dashboard, directive workflows and long-term notes.
- MIT license. No hosted MUSE service is required for the local continuity runtime.

## Verification boundaries

The 3.7 release adds full-scope handoff regressions for bounded reads, source and
scope drift, invalid proofs and concurrent writers. The 3.6 native Codex → Claude
Code → Codex exercise remains historical evidence; it does not certify the new
batch protocol. Synthetic fixtures are not native evidence. Missing source coverage stays incomplete;
Save, Bye and automatic compaction do not guarantee lossless memory or product
readiness. Native Windows and network filesystem durability are not certified.

## Documentation

- [Plain-language beginner guide](https://muse.mythslabs.ai/guide-en.html)
- [中文入门：让 AI 接着上次做](https://muse.mythslabs.ai/guide.html)
- [Beginner guide Markdown source](https://github.com/myths-labs/muse/blob/main/docs/QUICKSTART.md)
- [English usage and migration](https://github.com/myths-labs/muse/blob/main/docs/CONTINUITY.md)
- [中文使用与迁移说明](https://github.com/myths-labs/muse/blob/main/docs/CONTINUITY_CN.md)
- [Optional Jev tools and data boundaries](https://github.com/myths-labs/muse/blob/main/docs/OPTIONAL_TOOLS.md)
- [Jev 可选工具与数据边界](https://github.com/myths-labs/muse/blob/main/docs/OPTIONAL_TOOLS_CN.md)
- [Checkpoint capacity and evidence storage](https://github.com/myths-labs/muse/blob/main/skills/core/muse-commands/references/CHECKPOINT_CAPACITY.md)
- [3.7.1 release and validation scope](https://github.com/myths-labs/muse/releases/tag/v3.7.1)
- [Source and full documentation](https://github.com/myths-labs/muse)
- [Skill index](https://github.com/myths-labs/muse/blob/main/SKILL_INDEX.md)
- [Website](https://muse.mythslabs.ai)
- [Dashboard](https://muse.mythslabs.ai/dashboard.html)