agentkan

A local-first roadmap board for building with AI agents. Epics and tasks live as plain JSON and markdown in your repo; phases group execution and optional releases group shippable scope (v1, v2). A single index.html renders them as a drag-and-drop board you can edit on localhost. No server, no build step, no account, no lock-in.

It fixes one friction: when an AI keeps appending to a markdown to-do file, the file grows without bound and you lose track of what is actually yours to do. agentkan splits state (structured, movable) from prose (context for agents) and puts you in control of what is done.

npx agentkan init
npx agentkan serve
npx agentkan epic new "Billing integration"
npx agentkan validate

Pin locally with npm install -D agentkan.

How it works

Your agent shapes work, the CLI stubs epics, and the board is where you operate (drag, toggle tasks, archive). See how it works for the mental model, daily loop, and step-by-step scenarios.

Docs

• Getting started — Install, scaffold docs/board/, open the viewer, and wire validate into your workflow.
• How it works — Day-to-day usage: three surfaces (agent, CLI, board), scenarios, and what is configurable.
• Data model — Field reference for roadmap.json, epics, tasks, next.json, and board.tokens.json. Use when editing JSON or reviewing agent changes.
• Viewer — What the board UI does: drag-and-drop, in-board epic body editing, archiving, re-skinning via tokens, read-write when served.
• Skill & AI workflow — How to install the Agent Skill so Claude follows board conventions (propose vs dispose, interview, handoff).
• Distribution & publishing — GitHub setup, npm publish, local npm pack testing, CI.
• Origin & naming — Why the package is called agentkan and early project decisions.

Contributors working on this repo: AGENTS.md.

MIT licensed.