explaingit

mrlesk/backlog.md

★ 5,546TypeScriptAudience · vibe coderComplexity · 2/5LicenseSetup · easy

TLDR

A task manager that stores every task as a Markdown file in your Git repo, built specifically to hand tasks to AI coding assistants like Claude Code one at a time.

Mindmap

mindmap
  root((repo))
    What it does
      Markdown tasks
      Git versioned
      Kanban board
    AI Integration
      Claude Code
      MCP protocol
      CLI instruction files
    Use Cases
      Solo project backlog
      AI task handoff
      Terminal Kanban
    Access
      CLI tool
      Web browser UI
      npm or Homebrew
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

Things people build with this

USE CASE 1

Break a feature into small AI-readable tasks and have Claude Code work through them one at a time.

USE CASE 2

Manage a solo project backlog in plain Markdown files that live alongside your code in Git history.

USE CASE 3

Run a live Kanban board in the terminal or browser without signing up for any external service.

Tech stack

TypeScriptNode.jsMCP

Getting it running

Difficulty · easy Time to first run · 5min

Install via npm, bun, Homebrew, or Nix, no external services or accounts needed.

MIT license, use freely for any purpose including commercial projects with no restrictions beyond keeping the copyright notice.

In plain English

Backlog.md is a task management tool that stores all its data as plain Markdown files inside a Git repository. Every task is a .md file sitting in a folder called backlog/ within your project, which means tasks travel with the code, stay in version control history, and are readable in any text editor without a separate account or service. The tool is built around a workflow called spec-driven AI development. The idea is that AI coding assistants (such as Claude Code, Gemini CLI, or similar tools) work better when given small, clearly-written tasks with explicit acceptance criteria rather than open-ended instructions. Backlog.md provides the structure: you describe what you want, ask an AI agent to break it into tasks, review those tasks, then hand them to the agent one at a time. The tool connects to AI assistants either through MCP (a protocol many AI coding tools support) or through plain CLI instructions files like CLAUDE.md. For people who want to work without AI tools, Backlog.md also functions as a standalone task manager. The CLI lets you create tasks, edit them, list by status, and search by keyword. Running backlog board displays a live Kanban board in the terminal. Running backlog browser opens a web interface with a drag-and-drop Kanban board, task editing forms, and real-time sync back to the Markdown files. Tasks can have acceptance criteria checklists, a definition-of-done template that applies to every new task, and references to the source files they touch. Git is optional, the tool can run in a filesystem-only mode for non-code projects. Backlog.md installs via npm, bun, Homebrew, or Nix. It works on macOS, Linux, and Windows, has no external service dependencies, and is MIT-licensed.

Copy-paste prompts

Prompt 1
I'm using backlog.md with Claude Code. Create a new task titled 'Add user authentication' with JWT acceptance criteria and a definition-of-done checklist.
Prompt 2
List all my open backlog.md tasks and pick the highest-priority one to implement next based on its acceptance criteria.
Prompt 3
Using the backlog.md CLI, show me how to create tasks from a feature list and assign them acceptance criteria before handing them to an AI agent.
Prompt 4
I have a backlog.md task file. Help me split it into three smaller subtasks and create separate .md files for each in the backlog/ folder.
Open on GitHub → Explain another repo

← mrlesk on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.