What problem does it solve? Coordinating deferred work across multiple AI agents and sessions is error-prone: tasks get double-claimed, progress context is lost between sessions, and there is no honest record of what was tried, cancelled, or failed. This Skill provides a file-based task tracker where each task is one markdown file and its directory (todo/doing/done/failed) is its state, with git operations acting as the claim lock. ## Core Features & Use Cases - Maildir-style task pipeline: Tasks move forward through todo → doing → done directories via slash commands like /backlog add, /backlog take, /backlog advance, /backlog progress, /backlog cancel, /backlog fail, /backlog rescue, and /backlog retry, with an append-only timestamped event log in each file. - Multiple storage backends: Choose between maildir-git (fully git-tracked), maildir-shared (atomic cross-worktree claims via a shared git-common-dir), github-issues (tasks as GitHub Issues via gh), or jira (work items via Atlassian CLI acli). - Race-safe claiming and stale-claim rescue: Claiming uses git mv collisions or atomic file creation so racing agents collide instead of double-working, and timed-out claims can be rescued after a configurable timeout. - Roadmap and reflection: A backlog/ROADMAP.md layer captures intent, principles, and named priority arcs that tasks link to via arc: frontmatter. - Use Case: An agent team working across git worktrees runs /backlog take to auto-claim the highest-priority unblocked task, logs progress notes as it works, and advances the task to done/ with a PR link when finished. ## Quick Start Ask the agent to set up a backlog in the current repository with the maildir-git backend and add your first task, for example: "Run /backlog setup with the maildir-git backend, then add a task called fix-login-alignment."
