claude-ltm
LaserPhaser/claude-ltm
Memória persistente de decisões para projetos Claude Code
Ver no GitHubSkill que adiciona memória de longo prazo ao Claude, persistindo decisões e contexto do projeto em arquivos Markdown estruturados. Garante que decisões arquiteturais, mudanças de escopo e discussões técnicas não sejam perdidas entre sessões, mantendo um registro executável e consultável do histórico do projeto.
Instalação
npx convez-ai LaserPhaser/claude-ltm1 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
- Manter histórico de decisões arquiteturais em projetos longos
- Recuperar contexto e justificativas entre sessões diferentes
- Documentar mudanças de escopo e pivôs técnicos automaticamente
- Onboard novos membros do time com decisões e razões do projeto
Veredito

Solução pragmática para um problema real em projetos longos com Claude Code: perda de contexto histórico. A abordagem baseada em Markdown sem dependências externas é sólida, mas requer disciplina do usuário e instruções claras no CLAUDE.md para funcionar consistentemente. Útil para equipes que trabalham em sprints longos.
README
A tradução para o português deste README ainda está na fila.
claude-ltm
Long-term memory for Claude Code. Persist project decisions across sessions so context is never lost.
The Problem
During large workloads, the context and decisions made at the beginning of a project get lost as Claude Code sessions fill up and context windows rotate. Teams lose track of why choices were made, what was decided, and what changed.
How It Works
claude-ltm adds a persistent decision memory to your project — a structured Markdown file that Claude reads at every session start and updates as decisions are made.
Think of it like human memory:
- A main file (
.memory/DECISIONS.md) holds short decision checkpoints — what you'd recall off the top of your head - Detail files (
.memory/details/) hold deep-dive context for complex decisions — the memories you dig into when needed
Claude is instructed via CLAUDE.md to:
- Read the memory at session start
- Update it when significant decisions are made during work
- Compact before finishing — a final pass to capture anything missed
No hooks, no external services. Just Markdown files and behavioral instructions.
Install
1. Add the marketplace
Add this to your ~/.claude/settings.json:
{
"extraKnownMarketplaces": {
"claude-ltm-marketplace": {
"source": {
"source": "git",
"url": "https://github.com/LaserPhaser/claude-ltm.git"
}
}
}
}
2. Install the plugin
In any Claude Code session:
/plugin install claude-ltm
3. Initialize in your project
In each project where you want long-term memory:
/ltm:init
This creates .memory/DECISIONS.md and appends instructions to your CLAUDE.md.
Commands
/ltm:init
Initialize the memory system in your project. Creates .memory/ directory and adds behavioral instructions to CLAUDE.md.
/ltm:compact
Manually trigger a compact pass. Claude reviews the current session and extracts any undocumented decisions into .memory/DECISIONS.md.
Use this mid-session when you want to checkpoint your decisions.
/ltm:save
Explicitly save a specific decision:
/ltm:save chose Stripe over Paddle for payments because of better international coverage
Claude formats it and adds it to the appropriate section in the memory file.
Memory File Format
Main file (.memory/DECISIONS.md)
# Project Memory
> Auto-maintained by claude-ltm. Last updated: 2026-03-30
## Architecture
### REST API with Express.js [2026-03-28]
Chose REST over GraphQL because:
- Team has more REST experience
- Client is a simple SPA, no complex nested queries
- Easier to cache at the CDN level
### PostgreSQL for storage [2026-03-30]
Was: MongoDB. Switched because reporting features need JOINs
and the data turned out to be more relational than expected.
## Status
- **Focus**: User authentication flow
- **Blockers**: Waiting on payment provider API key
- **Open**: JWT vs session-based auth?
## Changelog
- 2026-03-30: Switched MongoDB → PostgreSQL
Rules:
- Each decision gets a
### headingwith date - Up to ~10 lines of context per entry
- Sections are free-form — Claude creates new ones as needed
- Superseded decisions are updated in-place with change notes
Detail files (.memory/details/)
For complex decisions that need more than 10 lines:
---
decision: API Architecture
date: 2026-03-28
status: ACTIVE
---
## Context
[Why this decision was needed]
## Decision
[What was chosen and why]
## Alternatives Considered
- Option B: [why rejected]
## Impact
- [Key consequences]
Uninstall
Per-project removal
Remove the LTM section from your project's CLAUDE.md — delete everything between <!-- BEGIN claude-ltm --> and <!-- END claude-ltm -->. Optionally delete .memory/.
Global removal
/plugin uninstall claude-ltm
Migrating from v0.1
If you previously installed via npx claude-ltm init:
- Delete these files from your project (the plugin system provides them globally now):
.claude/commands/ltm-init.md.claude/commands/ltm-compact.md.claude/commands/ltm-save.md
- Your
.memory/directory andCLAUDE.mdinstructions are unchanged — no migration needed - Install the plugin as described above
License
MIT