dsh-kanban

When running multiple AI agents through DeepSeek Harness, tasks tend to scatter across separate sessions with no single view to track the full lifecycle from creation and execution to review and merge. dsh-kanban links projects, git branches, and AI agents into one automated pipeline: once a task is created, an agent claims it and works in an isolated kanban/ branch, commits results, and a human reviewer merges back with --no-ff before the task branch is cleaned up — all visible right from the sidebar.

The plugin ships dual kanban and roadmap (Gantt) views backed by a complete task state machine (todo, running, review, approved, done, plus paused). Every task carries a full change log of the agent's final output, and you can append follow-up instructions through comments to resume the same agent session. The toolbar also integrates model and time scheduling, git branch selection, and path autocomplete for descriptions and comments. The sidebar is self-implemented without patching DSH internals, so the official workspaces, settings, and footer slots render as usual, and other sidebar plugins continue to work alongside it.

Ideal for solo developers or small teams who drive multiple AI agent tasks through DSH and want a visual pipeline that cuts down context switching. All state lives in the local DSH storage domain, git operations stay within the selected project, and there is no telemetry or external data reporting — privacy is guaranteed by architecture.