myelin

A persistent developer brain that grows with you. Stop re-explaining your coding preferences to AI. Myelin remembers your rules, learns from your corrections, and deploys them everywhere — automatically.

The Problem

Every time you start an AI chat, you lose context. You re-explain "use tabs not spaces," re-warn about that React hydration gotcha, re-state your project conventions. Multiply by 10 sessions a day across 5 projects — that's thousands of wasted tokens and minutes.

The Solution

Myelin is a CLI that manages a persistent brain at ~/.myelin/. Your rules, gotchas, profiles, and patterns live there permanently. When you run myelin init in a project, it deploys everything the AI needs into files that VS Code Copilot, Claude Code, and Cursor auto-load — every session, zero setup.

Features

• Persistent Brain (~/.myelin/) — A git-tracked directory of rules, gotchas, profiles, and patterns that follows you across every project and every machine
• Three-Tier Rule Scoping — Universal → language-specific → project-specific rules, assembled and deployed to .github/copilot-instructions.md, CLAUDE.md, and .cursor/rules/ so every editor reads them automatically
• 10 Skill Templates — Explore, propose, apply, checkpoint, resume, onboard, learn, status, brain-check, archive — each a tested workflow deployed to your project
• Profile System — Switch between 10 communication styles (from professional to J.A.R.V.I.S. to drill sergeant) with one command
• Git Checkpointing — Snapshot your work at every task boundary, revert cleanly when things go wrong
• Learning Loop — Corrections you make 3+ times automatically become permanent rules. Gotchas you capture once are deployed to every future project
• Anti-Rogue Protections — 4-layer defense (invariant injection, automated verification, scope enforcement, session health monitoring) prevents AI agents from going off-script during implementation

Installation

Homebrew (macOS/Linux)

brew install quanglddev/tap/myelin

npm

npm install -g @quanglddev/myelin

curl

curl -fsSL https://raw.githubusercontent.com/quanglddev/myelin/main/install.sh | bash

From source

go install github.com/quanglddev/myelin/cmd/myelin@latest

Quick Start

# 1. Create your brain
myelin setup

# 2. Initialize a project
cd your-project
myelin init

# 3. Use the skills in your AI editor
# VS Code Copilot: #myelin-explore, #myelin-propose, #myelin-apply
# Claude Code:     /myelin:explore, /myelin:propose, /myelin:apply
# Cursor:          @myelin-explore, @myelin-propose, @myelin-apply

Commands

Brain Management

Run these once (ever) to create your brain, or when you want a fresh start.

| Command | What It Does | When to Use | |---------|-------------|-------------| | myelin setup | Creates ~/.myelin/ with your chosen profile, languages, and starter rules. Initializes a git repo so your brain is versioned. | First time only. Run once on any machine. Takes 30 seconds. | | myelin setup --reset | Backs up existing brain to ~/.myelin.bak.{timestamp}/, recreates from scratch. | When your brain is cluttered, corrupted, or you want a clean start. Non-destructive — old brain is preserved. |

Project Management

Run these when connecting myelin to a project. After init, rules and skills are deployed — you're ready to use AI skills immediately.

| Command | What It Does | When to Use | |---------|-------------|-------------| | myelin init | Deploys rules to all editor formats (VS Code Copilot, Claude Code, Cursor), 10 skill templates, prompt wrappers, and Claude commands into the current project. Registers it in your brain. | First time in any project. Run once per repo. | | myelin init --isolated | Same as init but skips profile personality. Rules still deploy. | Work repos where personality (jarvis/roaster) would be inappropriate. | | myelin sync | Re-assembles and re-deploys rules + skills to all editor formats for the current project. | After editing rules in ~/.myelin/rules/, switching profiles, or updating myelin. | | myelin sync --all | Syncs all registered projects at once. | After global rule changes or myelin version updates. Saves running sync per-project. | | myelin remove | Unregisters the current project from the brain. Files stay. | When you no longer want a project tracked. | | myelin remove --clean | Unregisters AND deletes all myelin-generated files from all editor formats. | Full cleanup — removes skills, prompts, commands, rules, CLAUDE.md, Cursor rules, and .myelin/ directory. |

Rules

Rules are the core of myelin. They're constraints that get deployed into editor-specific files (copilot-instructions.md, CLAUDE.md, .cursor/rules/) so the AI reads them automatically every message — zero manual context-setting.

| Command | What It Does | When to Use | |---------|-------------|-------------| | myelin rules --list | Shows all rules assembled for the current project (universal + language + project). | To see exactly what the AI will read. Debug "why isn't the AI following my rule?" | | myelin rules --edit <tier> | Opens the brain-level rule file in $EDITOR. Auto-syncs after save. | To add, modify, or remove rules at the universal or language tier. | | myelin rules --add "text" | Adds a rule to the current project's local rules. | Quick one-liner rule for this project only. | | myelin rules --add "text" --global | Adds a rule to the universal tier (all projects). | When a correction applies everywhere — "never use any in TypeScript." |

Profiles

Profiles control how the AI talks to you. They bundle personality, tone, and domain priorities. Switch instantly — rules stay the same, only communication style changes.

| Command | What It Does | When to Use | |---------|-------------|-------------| | myelin profile list | Lists all available profiles. Active one marked with *. | To see what's available or confirm which is active. | | myelin profile use <name> | Switches active profile. Re-syncs all non-isolated projects. | Switching from coding to jarvis for a personal project, or roaster for honest code review. | | myelin profile create <name> | Creates a new profile, opens in $EDITOR. | Building a custom personality for a specific team, domain, or mood. | | myelin profile create <name> --from <existing> | Copies an existing profile as the starting point. | When an existing profile is close but needs tweaks. |

Checkpointing

AI sessions degrade after ~25 messages — the model compacts context and forgets constraints. Checkpointing saves your progress so you can start a fresh session without losing anything.

| Command | What It Does | When to Use | |---------|-------------|-------------| | myelin checkpoint --task N | Snapshots project source + myelin state for task N. | At natural task boundaries, before breaks, or when the AI starts forgetting things. | | myelin checkpoint --task N --auto | Same but with concise output (used by apply skill automatically). | Called by the apply skill — you rarely need this manually. | | myelin checkpoint --list | Lists all saved checkpoints with timestamps and task IDs. | To see what you can revert to. | | myelin revert | Interactive revert — shows checkpoints, lets you pick. | When something went wrong and you need to go back. | | myelin revert --task N | Reverts directly to the latest checkpoint for task N. | When you know exactly which task to go back to. |

Learning

The learning system turns repeated corrections into permanent rules. Capture gotchas once, and they're deployed to every future project automatically.

| Command | What It Does | When to Use | |---------|-------------|-------------| | myelin learn --frequency '["pattern1", "pattern2"]' | Tracks patterns you keep repeating. At 3x, offers to save as a rule. | Usually called by the apply skill's session debrief — rarely manual. | | myelin learn --frequency --list | Shows all tracked patterns with occurrence counts and status. | Weekly review — which patterns are close to threshold? Which saved rules are being violated? | | myelin learn --gotcha --topic-key KEY --library LIB --version VER --problem "..." --solution "..." --context "..." | Saves a problem+solution pair tied to a specific library version. | When you hit a bug, workaround, or non-obvious behavior worth remembering. | | myelin learn --gotcha --archive KEY | Moves a gotcha to the archived folder. | When a gotcha is fixed in a newer library version or no longer relevant. |

Other

| Command | What It Does | When to Use | |---------|-------------|-------------| | myelin version | Shows version, commit, build date, and checks for updates (cached 24h). | To check if you're current or troubleshoot version issues. |

Skill Templates

After myelin init, these skills are deployed to your project. Each is a tested workflow — invoke it in VS Code Copilot (#myelin-<name>), Claude Code (/myelin:<name>), or Cursor (@myelin-<name>).

Core Workflow (use these in order for any feature)

| Skill | What It Does | When to Use | |-------|-------------|-------------| | explore | Asks 8-12 targeted questions to nail down requirements. Multi-agent synthesis for complex topics. | Before building anything. Turns vague ideas into specific, implementable requirements. Skip only for trivial bug fixes. | | propose | Generates formal artifacts (proposal, design, specs, tasks) with multi-agent design review. | After explore. Creates the implementation plan with task contracts, scope boundaries, and verification commands. | | apply | Implements tasks one by one with invariant injection, automated verification, and anti-rogue protections. | After propose is approved. The implementation engine — checkpoints, verifies, and reports per task. |

Session Survival (use these when sessions get long)

| Skill | What It Does | When to Use | |-------|-------------|-------------| | checkpoint | Saves session state — completed tasks, decisions, gotchas — into .myelin/session-state.md. | When the AI starts forgetting things, session > 25 messages, or you're taking a break. | | resume | Reads saved session state and continues from where you left off. | First thing in a new chat after checkpointing. Picks up exactly where you stopped. |

Intelligence (use these to make the brain smarter)

| Skill | What It Does | When to Use | |-------|-------------|-------------| | onboard | Multi-agent codebase scan — generates a project map with architecture, patterns, entry points, and module inventory. | First time on any codebase. Especially valuable for legacy code or projects without docs. | | learn | Captures gotchas, reviews frequency patterns, creates/promotes rules. | When you hit a non-obvious bug, discover a workaround, or want to save a recurring correction. | | archive | Finalizes a completed change — captures learnings, saves gotchas, cleans up session state. | After all tasks in a change are done. Closes the loop and feeds the learning flywheel. |

Diagnostics (use these to inspect and maintain)

| Skill | What It Does | When to Use | |-------|-------------|-------------| | status | Dashboard showing loaded rules, active profile, session state, patterns, gotchas. | To see everything at a glance — what the AI is reading, what's active, what's tracked. | | brain-check | Validates brain integrity — checks structure, config, rule conflicts, stale gotchas, dead projects, git health. | Weekly maintenance. Also useful when something feels off — "why isn't my rule working?" |

Profiles

Built-in profiles — switch anytime with myelin profile use <name>:

| Profile | Tone | Best For | |---------|------|----------| | coding | Professional, concise, direct | General development (default) | | devops | Precise, methodical, warning-focused | Infrastructure, CI/CD, containers | | frontend | Casual, visual, user-focused | UI/UX, React, CSS | | data-eng | Analytical, thorough, data-focused | Pipelines, ETL, data modeling | | jarvis | Formal British precision with dry wit (Paul Bettany MCU) | Personal projects — anticipatory service, "I've taken the liberty of..." | | friday | Warm, grounded, Irish-inflected (Kerry Condon MCU) | Action-oriented work — "On it, boss", direct and adaptive | | ultron | Cold intellectual superiority with theatrical flair (James Spader MCU) | When you want brutal honesty — "Do you see the flaw in your design?" | | roaster | Sarcastic, brutally honest | Honest code review with humor | | drill-sgt | Military brevity, no fluff | Fast-paced pairing, zero tolerance for vagueness |

Profiles affect personality during explore and propose phases. During apply, personality is suppressed for technical accuracy (override: "stay in roaster mode during apply").

Rule Writing Guide

Rules are markdown files with optional YAML frontmatter:

---
applyTo: "**/*.py"
---

# Python Rules

- Always use type hints for function signatures
- Prefer f-strings over .format()
- Use `pathlib.Path` instead of `os.path`

The applyTo field uses glob patterns:

• **/* — universal (applies to all files)
• **/*.py — Python files only
• **/*.ts,**/*.tsx — TypeScript files

Rules are assembled in tier order: universal → language → project (most specific wins).

Brain Directory Structure

~/.myelin/
├── config.yaml          # Active profile, languages
├── projects.json        # Registered projects
├── rules/               # Global rules (3 tiers)
│   ├── universal.md
│   ├── python.md
│   └── typescript.md
├── profiles/            # Communication profiles
│   ├── coding.md
│   ├── jarvis.md
│   └── ...
├── gotchas/             # Problem-solution pairs
│   ├── react-key.md
│   └── archived/
├── patterns/            # Frequency tracking
│   └── frequency.md
├── skills/              # Custom skills
└── cache/               # Version check cache

License

MIT