kanban-md
A file-based, agents-first kanban board with a single-binary CLI and text interface, using atomic claim-and-move operations and installable agent skills so multiple AI agents can pick up tasks and work a shared board in parallel without clashing.
kanban-md stores every task as a plain markdown file rather than in a database or server, so a project's board works with git, grep, and ordinary editors as well as with agents and humans. Its pick-and-claim command atomically finds, claims, and moves the next available task, with claims that expire automatically, so several agents can work the same board without stepping on each other; a self-healing check also detects and repairs duplicate task IDs, filename and frontmatter mismatches, and ID-counter drift.
Installing its skills teaches an agent the expected workflow: claim a task, create a worktree, implement, test, commit, release the claim, and mark the task done, which can then be run in a single agent or across several running in parallel while a terminal UI shows live progress. Token-efficient compact output formats keep the tool cheap for an agent to call repeatedly.
It installs as a Go binary through Homebrew, and tickets can be created manually or by asking an agent to add one in natural language.