MDPM

Overview

A markdown-native project tracker built as a Claude Code plugin. Tasks are plain .md files with YAML frontmatter, and status is just a directory: backlog/, active/, done/. Zero dependencies, fully file-based, and built so AI coding agents can handle project management alongside actual engineering work.

The Problem

I kept context-switching between Claude Code and Jira just to track what I was working on. The AI agent is right there in the repo, but it can't manage tasks without an external tool. I wanted something that lives in the repo, works for both humans and AI agents, and optionally syncs to Jira or Wrike when stakeholders need visibility.

What I Built

• — Claude Code plugin with 15+ slash commands for task management, planning, and triage
• — Stdlib-only Python CLI that mirrors every slash command for scripting and inline use
• — Directory-as-status convention so ls tells you the state of the project
• — Automatic work log appended to each task file as an append-only audit trail
• — Planning command that breaks a feature into 3-8 tasks with dependency mapping
• — Optional Jira and Wrike sync via MCP servers
• — Local kanban board server with zero external dependencies

Key Decisions

Stdlib-only Python so there's no pip install, no venv, just clone and go. The file system is the database, which means standard Unix tools work out of the box. Slash commands and the CLI share a core library so behavior is always identical. Task files are never deleted, only moved between directories, so you never lose history.