Backlog.md is a markdown-native task manager and Kanban visualizer designed for managing project collaboration between humans and AI agents within Git repositories.
The tool addresses the challenge of reviewing large volumes of AI-generated code by structuring agent work around three review checkpoints: reviewing task specifications before implementation, reviewing implementation plans before code is written, and reviewing code organized as one task per pull request. Tasks are stored as plain markdown files in the repository, allowing completed work to remain as a permanent, legible record of what was attempted and why. The approach keeps diffs to a human-readable size and catches misunderstandings early when they are still single sentences rather than rebuilt features.
Backlog.md suits teams working with AI agents to generate code, particularly those who want to maintain meaningful human oversight throughout the development process. The tool works with various AI assistants including Claude Code, Gemini CLI, and others compatible with MCP or CLI interfaces. It provides acceptance criteria and definition of done checklists per task, milestone and dependency tracking for structuring larger efforts, and multiple interfaces including a terminal Kanban board, a local web UI with drag-and-drop editing, and fuzzy search across tasks. The project is self-contained and requires zero configuration to start using.
The project is actively maintained with regular updates to its feature set and capabilities. Development work on the tool itself is conducted through the same Backlog.md workflow it provides, with the full task ledger visible in the repository's backlog folder, demonstrating the tool's practical application in real development. The tool has been presented at multiple developer conferences, indicating ongoing engagement with the development community.
