# MUSE

A Markdown-based governance system for AI pair programming that adds role isolation, persistent dual-layer memory, sixty-six reusable skills and cross-role directives on top of a plain instructions file, with a Python runtime for guarded continuity.

MUSE extends the idea of a plain instructions file such as AGENTS.md or .cursorrules into a fuller system for AI pair programming: role isolation between different jobs the assistant does, persistent memory that survives across separate conversations, sixty-six reusable skills, a cross-role directive queue, and visual dashboards, all expressed in plain Markdown with a Python runtime that guards checkpoint updates. The core workflow is explicit: a user explains what they want, the assistant plans, works, checks its own work and records progress, and a small set of commands resumes earlier work, saves current progress, or formally wraps up a session.

The project addresses the problem that format specs alone cannot manage memory, roles or knowledge across sessions, so long conversations lose early context and new conversations start from nothing; MUSE's dual-layer memory and role system are meant to close that gap. Optional third-party tools for browsing, review and retrieval decisions are mentioned as compatible but are not bundled, and ordinary resume-and-save use does not require them; the project is written in Python.

Website: https://github.com/myths-labs/muse