lee-to/aif-handoff

Autonomous Kanban board where AI agents plan, implement, and review your tasks — fully hands-off.

AIF Handoff – Autonomous AI‑driven Kanban

What it is – A web‑based Kanban board that lets large‑language‑model agents plan, implement, and review code‑oriented tasks without human intervention. The system is built on the open‑source AI Factory framework and uses pluggable runtime adapters for Claude, OpenAI Codex, OpenRouter and OpenCode.

Core idea – A task moves through a fixed pipeline (Backlog → Planning → Implementing → Review → Done). At each stage a specialised AI sub‑agent (or a single “skill” function) generates the artefacts, runs quality checks, and can automatically loop back for re‑work until a convergence condition is met. When convergence fails, the task is handed off to a human for manual review.

Key Features (as described in the README)

  • Fully autonomous pipeline – create a task and the AI plans, writes code, and reviews it.
  • Kanban UI – drag‑and‑drop board with live updates via WebSocket.
  • Modular runtimes – swap Claude, Codex, OpenRouter or OpenCode per project or per task without changing orchestration logic.
  • Sub‑agent orchestration – dedicated coordinators for planning, implementation and review (plus security side‑cars).
  • Layer‑aware execution – implementation stage computes dependency layers and runs parallel workers when possible.
  • Self‑healing – heartbeat & watchdog automatically recover stuck agents.
  • Human‑in‑the‑loop options – approve plans, request changes, or run completely hands‑off.
  • Participants mode – optional local user accounts with role‑aware ownership of tasks.
  • GitHub integration – import issues, create isolated PRs, and require a human‑merged PR per task.
  • Model Context Protocol (MCP) sync – bidirectional task sync with other AIF tools.

Architecture at a glance

packages/
├─ shared/      # types, constants, logger
├─ runtime/     # registry & adapters (Claude, Codex, …)
├─ data/        # SQLite access (drizzle‑orm)
├─ api/         # Hono REST + WebSocket server (port 3009)
├─ web/         # React + Vite UI (port 5180)
└─ agent/       # coordinator + sub‑agent orchestration

The agent polls every 30 s, dispatches the appropriate sub‑agents, and tracks task liveness via heartbeats. Auto‑review can run in two strategies (full_re_review or closure_first) and stops when convergence fails, marking the task with manualReviewRequired=true.

Getting Started (quick‑start)

Without Docker

git clone https://github.com/lee-to/aif-handoff.git
cd aif-handoff
nvm use               # Node 20.19+ or 22.12+
cp .env.example .env
npm install
npm run init         # initial DB setup
npm run dev          # starts API, Web UI and background agent

If you switch Node versions, rebuild the native better-sqlite3 binary: npm rebuild better-sqlite3.

With Docker

git clone https://github.com/lee-to/aif-handoff.git && cd aif-handoff
cp .env.example .env
mkdir -p projects               # host directory for created projects
docker compose up --build      # starts API, UI, agent (+ optional MCP)
# Authenticate Claude inside the container
docker compose exec agent claude login
docker compose restart agent

Docker handles Codex SDK/CLI installation; you can override the version via CODEX_VERSION.

Runtime Providers

Provider Transport options
Claude (Anthropic) SDK (direct Agent SDK) – risk of violating Anthropic terms; safer CLI transport; API transport via ANTHROPIC_API_KEY.
Codex (OpenAI) SDK, CLI, App‑Server, API
OpenRouter API
OpenCode API
You can add custom adapters via the AIF_RUNTIME_MODULES env variable.

Execution Modes

Mode AGENT_USE_SUBAGENTS Behaviour
Subagents (true) Each pipeline stage runs through dedicated coordinator agents, iterating until quality criteria are met. Higher token usage, higher quality.
Skills (false – default) Stages are single AIF skill calls (/aif-plan, /aif-implement, etc.). Faster, lower token cost; optional Improve/Verify steps add extra checks.

Who might use this?

  • Teams that want to experiment with end‑to‑end AI‑generated code without writing boilerplate orchestration.
  • Researchers exploring autonomous software development pipelines.
  • Solo developers looking for a “set‑and‑forget” task board powered by LLMs.

License & Community

The repository is open‑source (MIT/Apache style typical for Lee‑to projects) and depends on the AI Factory framework, which is also open source. Issues, feature requests, and contributions are welcomed via GitHub.

Bottom line: AIF Handoff is a genuine, production‑grade project that demonstrates how to combine multiple LLM providers, modular runtime adapters, and a Kanban UI to create a hands‑off software development pipeline.

Related

  • Project
    Prorise-cool/Claude-Code-Multi-Agent

    Claude Code Multi‑Agent is an open‑source framework that gives Anthropic’s Claude Code project awareness and expert‑agent capabilities. It uses a Python Hook system and a locally‑run Ollama model to automatically detect the codebase type, analyse user intent, load 300+ domain‑specific “Skills” (backend, frontend, testing, security, etc.), and enforce automatic documentation (DEVELOPMENT.md, KNOWLEDGE.md, CHANGELOG.md). After cloning, installing Ollama, a model, and uv, you drop your project into the repo and Claude Code instantly knows its context, suggests appropriate tools, and can be driven with slash commands like `/backend‑specialist` or workflow commands such as `/kiro/spec`. Key points: - Zero‑config start; all configuration lives in `.claude/settings.json` and `prompts.json`. - Hooks run on session start, user prompt, tool use, etc., handling project detection, intent analysis, and doc updates. - Skills are defined by `SKILL.md` files and are auto‑discovered. - Works entirely locally (no external API), integrates with Git, and is MIT‑licensed. Ideal for developers or small teams who want a reproducible, on‑premise AI coding assistant.

  • Project
    xiaYuTian11/maskit

    Maskit is a local privacy‑gateway that automatically masks sensitive data in prompts sent to large‑model APIs and restores the original values in the responses, offering desktop, Docker, and browser‑extension deployments.

  • Project
    modu-ai/moai-cowork

    MoAI‑Cowork is an open‑source plugin suite that adds a set of Korean‑focused AI “co‑workers” (writer, marketer, designer, lawyer, etc.) to Claude Cowork and ChatGPT Work. A PM agent gathers the project goal, builds a workflow of the appropriate role‑agents, and runs the tasks. The repo provides manifests for both platforms, role‑specific plugins, and a full documentation site. Output is owned by the user and the software is Apache‑2.0 licensed.

  • Project
    narcotic-sh/senko

    A high-speed, accurate speaker diarization pipeline that identifies who spoke when in audio recordings, capable of processing one hour of audio in seconds.