create-agent

分类: 数据与AI | 上传者: open-ma | 下载: 0 | 版本: v1.0（最新）

帮助用户通过对话创建和配置 openma 托管代理。当用户说“创建代理”、“我需要一个代理......”、“为其设置代理”、“为我构建一个机器人”或描述要自动化的任务时触发。还可以从仪表板触发“使用 AI 创建”。当用户询问 openma 平台、它可以做什么、如何使用 CLI 或如何配置资源时也可以使用。 --- # openma Agent Creator ## 什么是 openma？ openma 是一个用于构建、部署和管理 AI 代理的开源平台。将其视为托管运行时 - 您定义代理（模型+系统提示+工具），平台处理沙盒执行、凭证管理和会话状态。 **您可以用它做什么：** - **为任何任务构建代理**：编码、研究、数据分析、客户支持、自动化 - **在沙箱中运行代理** - 每个会话都会获得一个带有文件系统、shell 和网络的独立容器 - **使用 OAuth 通过 MCP 服务器（GitHub、Slack、Linear、Notion 等）连接外部服务** - **使用任何 LLM** - Anthropic、OpenAI、DeepSeek 或任何 OpenAI 兼容提供商 - **从 ClawHub 安装社区技能**扩展代理功能 - **在保管库中安全地管理凭证** - 代理获得范围内的访问权限，秘密永远不会泄漏 - **协作** - 具有用于 CLI/SDK 集成的 API 密钥访问的多用户工作区 ## 创建代理 ### 流程 1. **了解目标** - 询问代理应该做什么。如果不清楚，就问一个问题：“主要任务是什么？”最多两轮，然后建造。 2. **选择型号** — 首先检查 `/v1/model_cards`。默认值： - 复杂/编码：`claude-opus-4-6` - 常规（默认）：`claude-sonnet-4-6` - 简单/快速：`claude-haiku-4-5-20251001` - OpenAI：`gpt-4o`、`o3` 3. **编写系统提示** — 具体、可操作、有界。不通用。 4. **选择工具** — 默认 `agent_toolset_20260401`（文件操作、bash、web）涵盖了大多数情况。 5. **创建：** ``` POST /v1/agents { "name", "model", "system", "tools": [{"type":"agent_toolset_20260401"}

更新日志: Source: GitHub https://github.com/open-ma/open-managed-agents

目录结构

当前层级: tree/main/skills/create-agent/

📄 SKILL.md 3.2 KB

SKILL.md

---
name: create-agent
description: >
  Help users create and configure openma managed agents through conversation.
  Trigger when users say "create an agent", "I need an agent that...", "set up
  an agent for", "build me a bot", or describe a task to automate. Also trigger
  for "Create with AI" from Dashboard. Also use when users ask about the openma
  platform, what it can do, how to use the CLI, or how to configure resources.
---

# openma Agent Creator

## What is openma?

openma is an open-source platform for building, deploying, and managing AI agents.
Think of it as a managed runtime — you define an agent (model + system prompt + tools),
the platform handles sandboxed execution, credential management, and session state.

**What you can do with it:**
- **Build agents** for any task: coding, research, data analysis, customer support, automation
- **Run agents in sandboxes** — each session gets an isolated container with file system, shell, and network
- **Connect external services** via MCP servers (GitHub, Slack, Linear, Notion, etc.) with OAuth
- **Use any LLM** — Anthropic, OpenAI, DeepSeek, or any OpenAI-compatible provider
- **Install community skills** from ClawHub to extend agent capabilities
- **Manage credentials** securely in vaults — agents get scoped access, secrets never leak
- **Collaborate** — multi-user workspace with API key access for CLI/SDK integration

## Creating an Agent

### Flow

1. **Understand the goal** — ask what the agent should do. If vague, one question:
   "What's the main task?" Two rounds max, then build.

2. **Pick the model** — check `/v1/model_cards` first. Defaults:
   - Complex/coding: `claude-opus-4-6`
   - General (default): `claude-sonnet-4-6`
   - Simple/fast: `claude-haiku-4-5-20251001`
   - OpenAI: `gpt-4o`, `o3`

3. **Write system prompt** — specific, actionable, bounded. Not generic.

4. **Select tools** — default `agent_toolset_20260401` (file ops, bash, web) covers most cases.

5. **Create:**
   ```
   POST /v1/agents
   { "name", "model", "system", "tools": [{"type":"agent_toolset_20260401"}] }
   ```

6. **Next steps** — offer to create session, configure skills, set up model card.

## Platform Quick Ref

Agents need a **session** to run. Sessions need an **environment** (sandbox).

| Resource | What it is |
|----------|-----------|
| Agent | Model + system prompt + tools config |
| Session | A conversation with an agent in a sandbox |
| Environment | Sandbox runtime (default works for most) |
| Model Card | API key + provider config for an LLM |
| Vault | Secure credential storage for MCP/CLI secrets |
| Skill | SKILL.md that gives agents domain expertise |
| API Key | Programmatic access token for CLI/SDK |

```
oma agents create <name>                    # create agent
oma sessions create --agent <id> --env <id> # start session
oma sessions message <id> <text>            # send message
oma models create --name <n> --model-id <id> --api-key <key>
oma keys create                             # generate API key
oma skills install <slug>                   # install from ClawHub
oma --help                                  # full command list
```

Model card providers: `ant`, `oai`, `ant-compatible`, `oai-compatible`.

登录后下载/点赞/收藏 ❤ 6 | ★ 0

create-agent

目录结构

SKILL.md

举报内容

提示