I built kanban-md because I wanted a simple local task tracker that works well for the agent loop: drop tasks in, run multiple agents in parallel, avoid collisions, and observe progress easily.

Tasks are just Markdown files (with YAML frontmatter) in a `kanban/` next to your code — no server, no DB, no API tokens. Simple, transparent, future-proof.

What makes it useful for multi-agent workflows:

- *Atomic `pick --claim`* so two agents don’t grab the same task.

- *Token-efficient `--compact` output* (one-line-per-task) for cheap polling in agent loops.

- *Skills included* -- just run `kanban-md skill install --global`; There is a skill for CLI use, and a skill for the development loop using the CLI (might need some additional work to be more general though, but works quite well)

- *Live TUI (`kanban-md-tui`)* for control ~~and dopamine hits~~.

I'd love feedback from anyone running multi-agent coding workflows (especially around claim semantics, dependencies, and what makes you feel in control).

I had a blast using it myself for the last few days.

Tech stack: Go, Cobra, Bubbletea (TUI), fsnotify (file watching). ~85% test coverage across unit + e2e tests. After developing webapps, the simplicity of testing CLI and TUI was so freeing.