hashgraph-online/awesome-codex-plugins / plugins/Colin4k1024/tsp/skills/agent-patterns-catalog
agent-patterns-catalog
VerifiedAI Agent 架构模式目录。提供 ReAct、Multi-Agent Conversation、Multi-Agent Workflow、 RAG Agent、Tool-Use Agent、Autonomous Agent 六种核心模式的参考架构、组件拓扑、 数据流、设计决策和框架无关的最佳实践。供 agent-dev-workshop Phase 2 调用。
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 2.7 KB |
| references/autonomous-agent.md | markdown | 4.7 KB |
| references/multi-agent-conversation.md | markdown | 3.4 KB |
| references/multi-agent-workflow.md | markdown | 3.8 KB |
| references/rag-agent.md | markdown | 4.4 KB |
| references/react-agent.md | markdown | 2.4 KB |
| references/tool-use-agent.md | markdown | 4.3 KB |
references/react-agent.md
View on GitHub# ReAct Agent 架构模式
## 核心思想
**Reasoning + Acting** — Agent 在每一步中先推理(Thought),然后决定行动(Action),
再根据行动结果观察(Observation),循环执行直到得出最终答案。
## 参考架构
```mermaid
graph TD
User[用户输入] --> Agent[Agent Core]
Agent -->|1. Thought| LLM[LLM 推理]
LLM -->|2. Action| Router{工具路由}
Router --> T1[Tool 1]
Router --> T2[Tool 2]
Router --> T3[Tool N]
T1 -->|3. Observation| Agent
T2 -->|3. Observation| Agent
T3 -->|3. Observation| Agent
Agent -->|4. Final Answer| User
Agent --> Memory[Memory Store]
```
## 执行循环
```
while not done and iterations < max_iterations:
thought = llm.think(prompt + history + observations)
if thought.is_final_answer:
return thought.answer
action = thought.selected_tool
observation = tools.execute(action.name, action.params)
history.append(thought, action, observation)
return "达到最大迭代次数,当前最佳结果: ..."
```
## 组件职责
| 组件 | 职责 | 关键配置 |
|------|------|---------|
| Agent Core | 主控循环、维护状态 | `max_iterations`, `timeout` |
| LLM Client | 推理、决策 | `model`, `temperature`, `max_tokens` |
| Tool Registry | 注册工具、分发调用 | 工具列表、参数验证 |
| Memory | 存储对话历史 | `type`, `window_size` |
## 适用场景
- 需要多步推理的通用任务(搜索 → 分析 → 汇总)
- 信息收集和整合
- 问答 + 操作(查询数据 → 执行修改)
- 客户支持(诊断问题 → 查找解决方案 → 执行操作)
## 设计要点
1. **Prompt 设计**:必须清晰描述可用工具、何时使用、输出格式
2. **终止条件**:明确定义什么是最终答案
3. **迭代上限**:防止无限循环,建议 5-15 次
4. **工具描述**:精确的工具描述是 ReAct 成功的关键
5. **错误恢复**:工具失败时 Agent 应能尝试替代方案
## 常见陷阱
| 陷阱 | 表现 | 解决方案 |
|------|------|---------|
| 工具描述模糊 | Agent 选错工具或传错参数 | 精确的 description + few-shot |
| 无限循环 | 反复调用相同工具 | 设置 max_iterations + 去重检测 |
| 过度推理 | 不必要地调用工具 | 在 prompt 中明确"能直接回答则直接回答" |
| 错误传播 | 一个工具失败导致后续全错 | 每步观察后评估是否继续 |
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 469k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 321k | Updated | Claude CodeCodex | Verified | Install |