DSH Plugins
返回列表
⚙️

dsh-taskboard-flow

工作流 更新于 2026.09.01

在终端中运行以下命令:

dsh plugin install markelayan/dsh-taskboard-flow

将以下提示词粘贴到 DeepSeek Harness 对话框中:

在 DeepSeek Harness 中执行 dsh plugin install markelayan/dsh-taskboard-flow 即可安装本插件(源码地址:https://github.com/markelayan/dsh-taskboard-flow),安装后需重启 dsh web 使配置生效。

插件介绍

dsh-taskboard gives you the kanban board, the cards, and a set of taskboard tools — but the cards do not move on their own. You push, they budge. Reviews need a human to notice. Multiple agents stay siloed. taskboard-flow is the missing flow layer: it watches the local ledger for card transitions and, the moment a card lands in a new column, fires the right action so the whole board runs without manual nudging.

Four capability pillars do the heavy lifting. First, an automated triage-and-execute pipeline: a backlog card spawns a triage session that decomposes it into model-pinned child tasks; a todo card executes through the real run API; an in_review card wakes the creator session live or resumes it after a restart. Second, cross-session communication: session_message delivers a message from any session to any other, adapting its strategy to the target state (idle vs mid-turn) so a human always sees it and the agent reads it at the next turn; the companion contacts tool turns raw session IDs into named aliases you can list, status-check, and message in one call. Third, decision and escalation routing: a comment prefixed with @orchestrator escalates to the project-bound orchestrator session, while [ASK] cards skip decomposition entirely and wake a designated answerer. Fourth, controllability: three-tier concurrency caps (global, per-project, per-column), three callback delivery modes (wake, notify, context), all configured in a local file-based patch — no database, no telemetry, no runtime dependencies.

It is built for individual developers and small teams already running multi-agent workflows on DeepSeek Harness. If you want to drop a work-order card into backlog and walk away, want several projects each with their own orchestrator yet sharing one messaging bus, or need an orchestrator to nudge a stuck executor with a single line — this is the piece that turns dsh-taskboard from a board you operate into a pipeline that operates itself.

使用场景

  • 把工单卡片放入 backlog 后自动分解、执行并回传审查
  • 多个代理会话通过命名联系人互相通知和协调
  • 用 @orchestrator 与 [ASK] 卡片实现人工决策和升级路由

适合人员

  • 已使用 DeepSeek Harness 管理多个代理会话的开发者
  • 希望减少看板手工推动和审查遗漏的小团队
  • 需要本地多代理工作流编排的自动化工程师