Token导航 LogoToken导航TokenDH.com
研究检索需要联网clawhub未标认证来源可访问clear审计通过

production-agent-builder生产 Agent 建设者

Agent Skill

production-agent-builder 用于查找、检索和筛选相关信息,适合在 OpenClaw 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

5,786

周安装

246

GitHub Stars

公开资料未说明

下载量

2,027
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:production-agent-builder(生产 Agent 建设者)
来源仓库:https://github.com/wholeinsoul/production-agent-builder
安装命令:
openclaw skills install production-agent-builder
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install production-agent-builder

简介

生产级 AI 代理构建框架,提供结构化八步开发流程。

  • 适用于设计新代理、规划架构与搭建自动化工作流。
  • 输出标准化模板与最佳实践指导文档。production-agent-builder 属于研究检索类 Skill,可作为该场景下的辅助能力补充。
  • 安装前需确认权限范围及是否触发联网、命令执行或文件读写。
  • 建议结合来源仓库与原始 README 进一步核验具体用法。

SKILL.md

name
ai-agent-builder
description
Structured 8-step framework for building production AI agents. Use when designing a new AI agent, planning agent architecture, building an automated workflow, or reviewing an existing agent's design. Covers task selection, step mapping, I/O specification, system prompt writing, memory design, safeguards, interface choice, and testing. Triggers on "build an agent", "design an agent", "agent architecture", "create an AI workflow", "production agent", "agent planning", "how to build an agent".

AI Agent Builder

Structured framework for building AI agents that work in production. Based on the Storm & Storm methodology.

When to Use

  • Designing a new AI agent from scratch
  • Planning architecture for an automated workflow
  • Reviewing or improving an existing agent's design
  • Teaching someone how to build agents

The 8-Step Process

Follow these steps in order. Each step has a clear goal and concrete deliverables.

Step 1: Choose a Task

Pick ONE painful, repeating workflow. Not "AI in general."

  • Must be repeatable (weekly+), follow steps, have clear I/O
  • Define success: "Given X, the agent should output Y so that Z happens."

Step 2: Map the Steps

Break the task into 4–7 steps: INPUT → ACTIONS → DECISION → OUTPUT

  • Classify each step: ⚖️ pure rules | 📖 heavy reading/writing | 🎯 judgement calls
  • Choose infrastructure (no-code vs dev-friendly)
  • You need: strong model + tool calling + basic logs

Step 3: Specify Inputs, Outputs & Tools

Treat the agent like an API, not a chatbot.

  • Define required input fields (text, file, URL, ID)
  • Define structured outputs (JSON/template the system can trust)
  • Attach tools: data (search/DB/CRM), action (email/Slack/tasks), orchestration (schedulers/webhooks/queues)

Step 4: Write the System Prompt

Create a clear role with: role definition, boundaries, style, 1–2 example conversations.

  • Use ReAct pattern: observe → think → act → reflect

Step 5: Add Memory

Three layers: conversation state, task memory, knowledge memory (vector store/file search).

  • Key question: "What does this agent need to remember for the next step to be smarter?"

Step 6: Add Safeguards

Gate high-risk actions (email, data changes, money) behind human approval.

  • Rules: never invent IDs, ask when ambiguous
  • Log every tool call and decision for audit

Step 7: Build the Interface

Match to where users work: chat, Slack command, button in app, or web form.

Step 8: Test

For each real example: watch the trace, score correctness + efficiency + time saved.

  • Tighten prompts/tools/rules where it fails. Iterate.

Detailed Reference

For expanded details on each step, including selection criteria, classification examples, tool categories, memory layer patterns, and a pre-launch checklist:

→ Read references/guide.md

Output Format

When using this framework to design an agent, produce a design document covering:

# Agent Design: [Name]

## Task & Success Criteria
[Step 1 output]

## Step Map
[Step 2 output — numbered steps with classifications]

## I/O Specification
[Step 3 output — inputs, outputs, tools]

## System Prompt
[Step 4 output — the actual prompt]

## Memory Architecture
[Step 5 output — which layers, what's stored]

## Safeguards
[Step 6 output — gated actions, rules, logging]

## Interface
[Step 7 output — chosen interface and why]

## Test Plan
[Step 8 output — example inputs, expected outputs, scoring criteria]

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

86.56%
按下载量换算1,755

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills