convez.ai
SkillAutomaçãoAvançado
beyondeye

AITasks

beyondeye/aitasks

IDE agentica no terminal, tasks em Markdown, git-native, zero infraestrutura

Ver no GitHub
BomPluginScore · de 0 a 100

IDE ágil baseada em terminal para desenvolvimento assistido por agentes de IA. Utiliza arquivos Markdown com frontmatter YAML armazenados no repositório git, eliminando necessidade de infraestrutura backend. Integra-se com Claude Code e outros agentes de codificação para otimizar transferência de intenção e velocidade de desenvolvimento.

Instalação

$npx convez-ai beyondeye/aitasks

1 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

Estrelas
9
Forks
1
Acompanhando
1
Saúde
42%
Último commit
há 1 dia
Versão
v0.42.0
Issues abertas
1
Contribuidores
2

Casos de uso

  • Orquestração de agentes de IA em projetos de desenvolvimento
  • Decomposição e paralelização de tarefas complexas com worktrees git
  • Code review assistido por IA com integração a PRs e issues
  • Memória de longo prazo para agentes vinculando código a tarefas

Veredito

Solução robusta para times que querem automatizar fluxos de desenvolvimento com agentes IA mantendo tudo versionado em git. Exige proficiência técnica mas elimina overhead de backends e databases.

README

A tradução para o português deste README ainda está na fila.


A full agentic IDE in your terminal — kanban board, code browser, agent monitoring, and AI-enhanced git workflows — integrated with Claude Code and all other supported coding agents via skills. Tasks are markdown files with YAML frontmatter stored in your repo alongside your code. No backend. No database. Just git.

Built for maximizing development speed 🚀 AND human-to-agent intent transfer efficiency 💬.

🎯 The Challenge

AI coding agents are proficient enough to handle real development tasks. The bottleneck is intent transfer — getting structured, contextual instructions to the agent without the human becoming the slowdown. aitasks optimizes both the context the agent sees and the speed at which a human can steer it.

💡 Core Philosophy

"Light Spec" engine: unlike rigid Spec-Driven Development, tasks here are living documents.

  • Raw Intent: a task starts as a simple Markdown file capturing the goal.
  • Iterative Refinement: an AI workflow refines task files in stages — expanding context, adding technical details, and verifying requirements — before code is written.

🏗️ Key Features & Architecture

  • 🖥️ Agentic IDE in your terminal — Board, Code Browser, Monitor, Brainstorm, and Settings TUIs in one tmux session via ait ide. Press j to hop between them.

  • 🧠 Long-term memory for agents — archived tasks and plans become queryable context; the Code Browser annotates each line back to the task/plan that introduced it.

  • 🔀 Tight git coupling, AI-enhanced — PR import/close, issue tracker integration, contribution flow, changelog generation, and AI-assisted reverts — all rooted in git commits and task metadata.

  • 🧩 Task decomposition & parallelism — auto-explode complex tasks into child tasks; sibling context propagates via archived plans; git worktrees + atomic locks for parallel agent work.

  • 🔍 AI-enhanced code review — per-language review guides, batched multi-file reviews producing follow-up tasks, QA workflow with test-coverage analysis.

  • 🤖 Multi-agent support with verified scores — unified codeagent wrapper over Claude Code and all other supported coding agents; per-model/per-operation scores accumulated from user feedback.

  • Dual-Mode CLI — interactive mode for humans (optimized for flow, no context switching) and batch mode for agents (programmatic task/status updates).

  • Battle tested — actively developed and used in real projects.

  • Fully customizable workflow — scripts and skills live in your project repo; modify them for your needs and contribute back via /aitask-contribute. See the Contribute and Manage Contributions workflow.

🖥️ Platform Support

PlatformStatusNotes
Arch LinuxFully supportedPrimary development platform
Ubuntu/DebianFully supportedIncludes Pop!_OS, Linux Mint, Elementary
Fedora/RHELFully supportedIncludes CentOS, Rocky, Alma
macOSFully supportedRequires Homebrew for bash 5 and coreutils (auto-installed by ait setup)
Windows (WSL)Fully supportedVia WSL with Ubuntu/Debian (see Windows guide)

⚡ Quick Install

Pick your platform:

PlatformInstall command
macOSbrew install beyondeye/aitasks/aitasks — see the Homebrew guide
Arch / Manjaro (AUR)yay -S aitasks or paru -S aitasks — see the AUR guide
Debian / Ubuntu / WSLDownload the latest .deb from Releases and sudo apt install ./aitasks_*.deb — see the Debian/Ubuntu guide
Fedora / RHEL / Rocky / AlmaDownload the latest .rpm from Releases and sudo dnf install ./aitasks-*.noarch.rpm — see the Fedora guide
Other (any POSIX)curl -fsSL https://raw.githubusercontent.com/beyondeye/aitasks/main/install.sh | bash

All install methods drop a single ait command on your $PATH — the global shim (~3 KB). The shim downloads the framework on demand when you run ait setup in your project, so the package itself stays tiny and you do not need to re-install it to get framework updates. See the installation guide for the full picture, and the packaging distribution status & roadmap for current limitations and how each channel might evolve toward more official repos.

After installing, cd into your project (the git repository root) and run ait setup to bootstrap the framework.

Windows users: Run from a WSL shell, not PowerShell. See the Windows/WSL guide.

Upgrade an existing installation:

ait upgrade latest
ait setup

Already have the global ait shim? Once any install method has placed ait on your $PATH, you can bootstrap aitasks in any new project directory by simply running:

cd /path/to/new-project
ait setup

The shim detects that no aitasks project exists, downloads the latest release, installs it, and then runs the full setup — all in one command.

📦 What Gets Installed

Per-project files (committed to your repo):

  • ait — CLI dispatcher script
  • .aitask-scripts/ — Framework scripts (task management, board, stats, etc.)
  • .claude/skills/aitask-* — Claude Code skill definitions
  • aitasks/ — Task data directory (auto-created)
  • aiplans/ — Implementation plans directory (auto-created)

Global dependencies (installed once per machine via install.sh and ait setup):

  • CLI tools: fzf, gh (for GitHub), glab (for GitLab), or bkt (for Bitbucket), jq, git
  • Python venv at ~/.aitask/venv/ with textual, pyyaml, linkify-it-py
  • Global ait shim at ~/.local/bin/ait
  • Claude Code permissions in .claude/settings.local.json (see Claude Code Permissions)

📖 Documentation

Documentation Website — Browse the full documentation online.

  • Overview — The challenge aitasks addresses, its core philosophy, and key features of the agentic IDE.

  • Installation — Quick install, platform support, setup, and git remote authentication.

  • Getting Started — First-time walkthrough from install to completing your first task.

  • Concepts — What each building block is and why it exists: tasks, plans, parent/child, folded tasks, review guides, execution profiles, verified scores, agent attribution, locks, and the IDE model.

  • TUI Applications — The terminal IDE: Monitor, Minimonitor, Board, Code Browser, Settings, and Brainstorm — hop between them with a single keystroke via ait ide.

  • Workflow Guides — End-to-end guides for common usage patterns: capturing ideas fast, tmux IDE, complex task decomposition, parallel development, code review, QA, PR workflow, and more.

  • Code Agent Skills — Reference for /aitask-pick, /aitask-explore, /aitask-create, and other skill integrations across Claude Code and all other supported coding agents.

  • Command Reference — Complete CLI reference for all ait subcommands.

  • Development Guide — Architecture overview, directory layout, library scripts, and release process.

📄 License

This project is licensed under the Apache License 2.0 with the Commons Clause condition.

What this means: ✅ You can: Use, copy, and modify the code for free, with an explicit patent grant from contributors.

✅ You can: Use aitasks as a library to power your own commercial products or SaaS applications.

❌ You cannot: Sell aitasks itself, or a derivative version of it, as a standalone product or service (e.g., selling a "Pro" version of the library or a managed aitasks hosting service) without prior written consent.

For the full legal text, please see the LICENSE file.

Ver o original no GitHub

Bom para

Relacionados