Armadra

使用

协调者与工作流

用 ama 拆分任务、组织团队,把稳定的协作存成可重跑的模板。

协调者 ama

ama 是 Armadra 自带的协调 Agent。给它一个目标,它把子任务落成画布上的 Agent 节点、分派下去、等待并收齐结果。ama 节点头部的「N 成员」打开分派抽屉,列出各成员任务的状态与汇总;失败的任务可以重试。

成员遇到权限请求时,ama 只报告「被阻塞」,不替你回答。

团队编排

由 Agent 发起的编排支持几种形状:

  • 并行:多个成员同时做不同部分;
  • 流水线:上一个完成后下一个开始(open-agent --after);
  • 汇总:多个成员的结果交给一个节点合并。

每个成员可以带一条自己的 Git worktree,互不干扰。

工作流

当一种协作方式稳定下来,可以把它存成工作流模板:

  1. 协调者交一份草案;
  2. 你在工作面板的「工作流」页确认成模板;
  3. 之后填参数起跑,或在自动化里定时运行。

每次运行在画布上建一个 Frame 与各角色节点,关卡处停下等你确认;运行记录可以回看与对比。

什么时候用工作流

开放任务的路径会随新发现变化,适合先在画布上探索;步骤明确、需要反复执行的流程,再沉淀成工作流。两种方式可以结合。

Guides

Coordinator and workflows

Use ama to split tasks and organize a team, and save a settled collaboration as a rerunnable template.

The ama coordinator

ama is the coordinator agent that ships with Armadra. Give it a goal and it turns subtasks into agent nodes on the canvas, assigns them, waits and gathers the results. The “N members” badge in the ama node header opens the dispatch drawer with each member’s task status and summary; failed tasks can be retried.

When a member hits a permission request, ama only reports it as blocked; it never answers for you.

Team shapes

Agent-initiated orchestration supports several shapes:

  • Parallel: members work on different parts at the same time;
  • Pipeline: the next starts when the previous finishes (open-agent --after);
  • Fan-in: several members’ results go to one node to merge.

Each member can work on its own Git worktree so they don’t step on each other.

Workflows

When a way of collaborating settles, save it as a workflow template:

  1. The coordinator proposes a draft;
  2. You confirm it as a template on the Workflows page of the work panel;
  3. Run it later with parameters, or on a schedule from automation.

Each run creates a frame and the role nodes on the canvas and pauses at gates for your confirmation; runs can be reviewed and compared.

When to use a workflow

Open-ended tasks change course as you learn, so explore them on the canvas first; turn steps that are clear and repeated into a workflow. The two approaches combine well.

在 GitHub 上编辑本页 Edit this page on GitHub