m_AI_l — Mensageria Assíncrona Entre Threads Claude
playtosor/m_AI_l
Memória compartilhada entre threads Claude — zero setup, só Markdown
Ver no GitHubSistema de coordenação leve para projetos multi-thread com Claude, usando apenas Markdown. Cria uma camada de memória compartilhada onde diferentes conversas trocam mensagens, decisões e estado do projeto sem perder contexto. Ideal para trabalhos longos onde threads atingem limite de contexto e precisam passar a bola para sucessores sem reiniciar tudo.
Instalação
npx convez-ai playtosor/m_AI_l1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Projetos longos que atravessam múltiplas conversas Claude
- Coordenação entre assistentes especializados (escrita, código, pesquisa) sem copiar-colar manual
- Handoff automático quando uma thread satura — próxima thread retoma exatamente do ponto deixado
Veredito

m_AI_l resolve um problema real em workflows multi-thread: evita reiniciar contexto e manter histórico manualmente. Solução minimalista e pragmática, mas exige disciplina do usuário em manter estrutura Markdown consistente — não é automático.
README
A tradução para o português deste README ainda está na fila.
m_AI_l v1.1
Asynchronous messaging for human–AI collaborative projects — built entirely in Markdown.
The problem
When you work with Claude across multiple conversations, each thread is an island.
Your coding assistant doesn't know what your writing assistant decided. Your research thread ignores what your planning thread concluded. You are the only link — manually copy-pasting context between sessions, rebuilding the same shared understanding over and over. And when a thread hits its context limit, you're back to square one.
This doesn't scale.
The solution
m_AI_l is a lightweight coordination system for multi-thread, human–AI projects.
No database. No custom server. Just a handful of Markdown files that act as a shared memory layer between your Claude conversations.
Each thread gets a mailbox. You drop messages in. Claude reads, acts, and archives. Everyone stays in sync — asynchronously, reliably, without you having to repeat yourself. And when a thread reaches its limit, it hands off cleanly to its successor. Nothing gets dropped.
Never lose a thread
Claude conversations have a context limit. When a thread saturates or gets interrupted, the usual outcome is: you start over, you re-explain, you lose momentum.
m_AI_l handles this differently. When a thread is nearing its limit, it prepares a structured handover — decisions made, state of play, next steps — and deposits it in its successor's mailbox. The new thread reads it via MAIL! and picks up exactly where the previous one left off.
To the rest of the team, nothing happened.
This makes long-running projects genuinely sustainable. Not just "Claude with better memory" — but a system where continuity is built in, and nothing gets dropped.
Who is this for?
m_AI_l is designed for non-developers who use Claude for complex, ongoing projects and feel the friction of starting from scratch every conversation.
If you've ever thought "I wish Claude remembered what we decided last week" — this is for you.
Tokens running out mid-session? Drop a quick message to the relevant thread before you lose your train of thought. It'll be waiting when you come back.
Developers working with Claude Code or agent frameworks will likely prefer more powerful tooling. This system optimizes for simplicity and zero setup.
Already using Claude's built-in memory? Good. Memory knows your preferences — m_AI_l coordinates your projects. They work better together.
How it works
common/
├── roster.md ← Team directory: who's here, what role, what mailbox
├── mailbox_HUMAN.md ← Your inbox
├── mailbox_ARCHI.md ← Claude thread "ARCHI" inbox
├── mailbox_DEV.md ← Claude thread "DEV" inbox
└── skills/ ← Behaviour instructions for Claude
├── mailbox_read.md
├── mailbox_write.md
└── ...
Every message follows a simple standard format:
SENDER // YYYYMMDD-HHMM // PRIORITY // SUBJECT // Body
That's it. No tooling required. Works with Claude Projects, local files, or any folder you can share between tabs.
The skills
m_AI_l comes with a set of Claude skills you install once and use everywhere:
| Skill | What it does |
|---|---|
mailbox_read | Claude reads its inbox, acts on messages, archives silently |
mailbox_write | Claude deposits messages to any team member's mailbox |
onboarding | Invites and integrates any new team member — Claude thread, external agent, or human |
briefing | Self-integration for an existing thread joining the system mid-project |
renew | Replaces a saturating or crashed thread transparently — same handle, handover via mailbox |
aro | One-shot setup agent — configures your whole installation interactively |
prune | Garbage collection — moves old archives to a pruned/ folder, project by project |
m_AI_l_help | Full command reference, readable by both humans and Claude |
claude_web_read | Lets a guest claude.ai session read a mailbox without archiving |
claude_web_write | Lets a guest claude.ai session deposit a message as CLAUDE_WEB |
Available commands
| Command | What it does |
|---|---|
MAIL! | Read and archive all pending messages |
MSG? | Count pending messages without reading them |
SEND! | Reflect and send relevant updates to the team |
SEND! HANDLE | Send a message to a specific thread |
SEND? | Propose a message for your validation before sending |
RENEW? | Check thread saturation level |
RENEW! | Replace this thread with a fresh one, handover included |
PRUNE! | Launch a cleanup of old messaging archives across all projects |
HELP! | Display the full command reference |
INVITE WRITE | Generate a startup prompt for a guest write session |
INVITE READ | Generate a startup prompt for a guest read session |
Getting started
Requires Claude Desktop — m_AI_l uses the filesystem MCP to read and write local files. The web version of Claude is not supported.
New to m_AI_l? Read
setup.mdfirst — it covers prerequisites and installation.
Run ARO — the one-shot setup agent:
Execute C:\path\to\common\skills\aro.md
ARO detects your language, walks you through configuration, and generates your first conversation startup prompt.
Full setup guide → setup.md
Known limitations
- Archive files grow over time — use
PRUNE!periodically to move old entries to apruned/folder and keep the system lean.
Origin
m_AI_l was built to coordinate the team that built it — from day one.
Every design decision, every skill update, every handover between threads went through the system itself. The architecture you're reading about is the same one that was used to write it.
Developed entirely with Claude's free tier — because good systems shouldn't require premium access to exist.
License
Licensed under CC BY 4.0 — use it, adapt it, build on it. Just keep the credit.
A project by a one-person multinational. 🌍