Orthogonal Agent Framework

正交性Agent框架 - 模块化AI Agent设计系统。将Agent能力拆分为规划、执行、记忆、评估四个正交维度，实现可组合、可替换、可测试、可扩展的Agent架构。

Security Scan

An instruction-only, modular agent design framework whose declared requirements and runtime instructions are consistent with its stated purpose and do not request extraneous credentials or install actions.

Purpose & Capability

Name/description (orthogonal modular agent framework) match the files and templates provided. Required binary (bash) is minimal and reasonable for templating or filesystem operations described. No unrelated credentials, binaries, or config paths are requested.

Instruction Scope

SKILL.md describes creating/maintaining local files (plan.md, execution.md, memory.md, evaluation.md) and using templates — this is appropriate for the framework. There are no instructions to read unrelated system files, access environment secrets, or send data to external endpoints. Note: the framework implies persistent local storage of 'memory' which could capture sensitive user data if used that way; the skill does not provide retention/erase/encryption guidance.

Install Mechanism

No install spec and no code to download or execute. Instruction-only distribution is lowest-risk and matches the content (just templates and docs).

Credentials

No required environment variables, credentials, or config paths. The absence of secrets is proportionate to a documentation/template-only framework.

Persistence & Privilege

always: false and the skill is user-invocable (normal). The framework expects to create and persist template files in the agent/workspace; this is expected but means local data (including any sensitive memory entries) may be stored—users should be aware of where files are written and how long they are retained.

Assessment

This skill is an instruction-only template for structuring agent work (planning, execution, memory, evaluation). It does not request credentials or download code, which reduces risk. Before using it: (1) verify where the agent will write the generated files (workspace path) and whether those files may contain sensitive information; (2) if you will store secrets or private data in the 'memory' files, ensure you have retention/erasure policies or encryption in place; (3) because the source/homepage are not provided, consider testing the skill in a sandboxed environment first and review templates to confirm they meet your operational and privacy requirements.

正交性Agent框架 - 模块化AI Agent设计系统

基于Manus核心设计理念，将AI Agent拆分为四个正交维度：

• 规划（Planning） - 任务拆解、目标设定
• 执行（Execution） - 工具调用、操作执行
• 记忆（Memory） - 上下文保持、长期存储
• 评估（Evaluation） - 结果检验、反馈改进

核心理念

什么是正交性？

正交性（Orthogonality）指各模块相互独立，改变一个不影响另一个。

┌─────────────────────────────────────────────────┐
│                  Agent System                     │
├─────────────┬─────────────┬─────────────┬────────┤
│  Planning   │ Execution   │   Memory    │   Eval │
│    模块      │    模块      │    模块      │   模块  │
└─────────────┴─────────────┴─────────────┴────────┘
       ↕            ↕            ↕            ↕
   可独立       可独立       可独立       可独立
   升级/       替换/       替换/       替换/
   测试        测试        测试        测试

优势

优势	说明
🔧 可组合	每个维度可独立升级
🔄 可替换	随时换掉某个模块
🧪 可测试	单独测试每个维度
📈 可扩展	新能力只需加新模块

使用方法

快速开始

使用正交性框架帮我规划一个复杂任务

框架文件

框架会自动创建以下文件管理任务：

文件	作用
plan.md	任务规划和拆解
execution.md	执行记录和步骤
memory.md	上下文和记忆
evaluation.md	结果评估和反馈

命令示例

命令	作用
正交性框架：规划 [任务]	拆解任务为步骤
正交性框架：执行 [步骤]	执行具体操作
正交性框架：记忆 [内容]	存储重要信息
正交性框架：评估 [结果]	检验并反馈
正交性框架：状态	查看当前各模块状态

适用场景

✅ 适合

• 复杂多步骤任务
• 需要长期记忆的任务
• 需要评估/反思的任务
• 构建自定义Agent系统

❌ 不适合

• 简单对话任务
• 一次性任务
• 固定流程

技术实现

模块接口

每个正交模块遵循统一接口：

Planning:
  - plan(task) → steps[]
  - update_plan(progress)
  - replan(feedback)

Execution:
  - execute(step) → result
  - tool_call(tool, args)
  - rollback(step)

Memory:
  - store(key, value)
  - retrieve(key)
  - forget(key)
  - summarize()

Evaluation:
  - evaluate(result) → score
  - feedback(issues) → improvements
  - validate(criteria)

状态管理

# agent-state.yaml
planning:
  current_task: "..."
  steps: [...]
  completed: [...]

execution:
  active_step: 1
  results: {...}

memory:
  short_term: [...]
  long_term: {...}

evaluation:
  scores: {...}
  feedback: [...]

扩展性

添加新模块

在框架中集成新能力：

modules:
  - planning     # 已有
  - execution    # 已有
  - memory       # 已有
  - evaluation   # 已有
  - creativity   # 新增 - 创意生成
  - research     # 新增 - 信息收集

替换模块

替换现有模块不影响其他：

# 替换记忆模块
memory:
  type: "vector_store"  # 从文件切换到向量数据库
  config: {...}

示例

完整任务流程

用户: 帮我研究AI Agent的发展趋势

正交性框架:
1. Planning → 拆解为5个子任务
2. Execution → 搜索信息
3. Memory → 存储重要发现
4. Evaluation → 评估信息质量
5. 循环2-4直到完成
6. 输出最终报告

文件结构

orthogonal-agent-framework/
├── SKILL.md
├── _meta.json
├── references/
│   ├── design-principles.md
│   └── module-interfaces.md
└── templates/
    ├── plan.md
    ├── execution.md
    ├── memory.md
    └── evaluation.md

---

基于Manus正交性设计理念构建