meow-dsh-task

DSH sessions do the work but rarely leave a trace: a task is claimed mid-conversation yet never recorded, a checkpoint is never persisted, and the next person picks it up by reconstructing context from chat logs. meow-dsh-task closes that gap. It persists tasks, development plans, checkpoints, reviews, issues, and handoff checklists as a JSON truth layer under .dsh-task/, then auto-renders them into human-readable Markdown under docs/tasks/. Everything ships with the repository via git, surviving across sessions, machines, and team handovers.

On the model side, fifteen meow_dsh_task_* tools cover the full task lifecycle: initialization, concurrency-guarded claiming, status transitions, checkpoint appends, in-task todo lists, review milestones, issue reporting and resolution, handoff writing, and plan CRUD. Five system-prompt sections (bootstrap, startup, exec, complete, review) are injected in order so the agent acts at the right moment with the right discipline. On the human side, a /meow-dsh-task command group and an embedded /meow-task web panel (opened via the Task button on the input bar) let people browse tasks, switch workspaces, and tune per-workspace settings like taskRoot. The entire plugin is a pure Cordis injection: zero DSH source modifications, fully offline, no telemetry, and the runtime depends only on Node built-in modules.

It is built for individuals and small teams already running agent workflows in the DSH web profile who want git-shared, structured task state without spinning up Jira or Linear. If your workflow fits the pattern of JSON-on-disk plus rendered Markdown in the repo, meow-dsh-task gives you a lightweight, zero-infrastructure task tracking layer out of the box. Note: TUI and CLI profiles are not supported.