Almanac
navasmo/almanac
Contexto unificado de estúdio: Linear + código + biblioteca em um comando
Ver no GitHubCamada de contexto unificada para assistentes de IA em estúdios de desenvolvimento. Integra status de projetos (Linear), varredura de código local, biblioteca de código reutilizável e regras de desenvolvimento em um único contexto acessível via CLI e Agent Skill para Claude.
Instalação
npx convez-ai navasmo/almanac1 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
- Fornecer ao assistente IA visão completa do projeto e histórico de código
- Centralizar regras de desenvolvimento e padrões reutilizáveis do estúdio
- Reduzir tempo de context-setup em cada sessão de trabalho com IA
- Gerenciar múltiplos projetos com contexto sincronizado
Veredito

Ferramenta bem pensada para estúdios solo/pequenos que trabalham com IA. Arquitetura sensata (CLI + Skill antes de MCP), mas ainda em fase inicial (P2 de 7). Vale acompanhar — resolve um problema real de context fragmentation.
README
A tradução para o português deste README ainda está na fila.
Almanac
Studio context for your AI. Almanac gives your AI assistant a single, trustworthy view of everything you're building — every project's status (from Linear), its code reality (from a local scan), your reusable code library, and your coding rules — fused into one context layer.
Built as a core library + almanac CLI + an Agent Skill, packaged as a Claude Code plugin. Self-hosted. An MCP adapter is an optional add for cross-client reach.
🚧 Building in public. Early development — see the milestones. Follow along.
Why
A solo studio juggling a dozen projects has its context scattered: status in Linear, code on disk, reusable patterns across repos. Every AI session starts from zero. Almanac fuses those sources so one command — almanac context <project> — tells your assistant the stack, the open issues, the design tokens, the applicable rules, and any related snippets you've already written.
Architecture
core library (Postgres: Linear pull · code scan · library · rules · security)
├── almanac CLI ← primary surface; your agent calls it via the shell
├── Agent Skill ← teaches the agent when/how to use it (portable SKILL.md)
├── plugin ← one-command install for Claude Code
├── dashboard ← a visual view of the studio (Next.js)
└── MCP adapter ← optional, for cross-client reach (later)
Why not an MCP server first? For a local, single-user tool, a CLI the agent calls via the shell plus a progressive-disclosure Skill is dramatically cheaper on context tokens and just as portable now that SKILL.md is an open standard. MCP stays the standard for connectivity — here it's a thin optional adapter.
Status
| Phase | |
|---|---|
| P0 | Bootstrap + security |
| P1 | Database core (Drizzle / Postgres) |
| P2 | Core + CLI + project context |
| P3 | Linear integration |
| P4 | Library + rules |
| P5 | Security scanner |
| P6 | Dashboard + deploy |
| P7 | Public polish |
Develop
pnpm install
pnpm verify # typecheck · lint · test · build
Requires Node ≥ 22 and pnpm. Copy .env.example to .env.local before running anything that touches the database.
Licence
MIT © Navas Moideen (NMO Digital)