dsh-task-kanban

When you drive an AI agent through DSH Web, a single requirement typically passes through several conversational stages (clarification, multi-phase implementation, and merge) yet the web UI offers no single panel to track all of them end to end. dsh-task-kanban closes that gap by bringing the full requirement-to-plan-to-implementation-to-merge pipeline into a sidebar view, with five swimlanes (Backlog, Queue, Running, Done, Merged) that let you see at a glance exactly where every card sits.

Create a task and the plugin immediately launches a refinement session. You can prefix your input with a slash to pick a Skill (e.g. /grill-me); the agent will probe, question, and shape the requirement following that Skill's instructions until it is concrete enough to be written back as a phased plan. Once a plan enters the queue, each phase runs in its own session inside a dedicated git worktree, strictly serial within one workspace. When every phase is done, the plugin merges the branch into main or master and cleans up the worktree automatically. Each card displays the model, reasoning level, plan badge, and phase, with stop, retry, and delete actions right on it.

This plugin is a good fit if you routinely use DSH Web to drive coding agents and want structured visual feedback on task progress, non-interfering serial execution across phases, and hands-off merging. The configuration surface stays small: a global worker cap, default models and reasoning levels for refinement versus implementation sessions, and an agent preset, all adjustable from the sidebar settings section.