A modular AI Agent design system that decouples agent capabilities into four independent dimensions for maximum flexibility and reliability.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install orthogonal-agent-framework
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install orthogonal-agent-framework using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Orthogonal Agent Framework?
The Orthogonal Agent Framework is a sophisticated design system inspired by the Manus core philosophy. It breaks down AI Agent architecture into four distinct, independent (orthogonal) dimensions: Planning, Execution, Memory, and Evaluation. This modularity ensures that changes to one component do not negatively impact others, allowing for cleaner development and more robust AI behaviors within the Openclaw Skills ecosystem.
By adopting this framework, developers can move away from monolithic prompt chains and toward a system where every part of the agent's logic is composable, replaceable, and testable. Whether you are building a research assistant or a complex automation tool, this framework provides the architectural rigor needed for production-grade AI agents.
Orthogonal Agent Framework Use Cases
- Orchestrating complex, multi-step tasks that require structured planning and sub-goal decomposition.
- Developing agents that need persistent long-term memory and context retention across sessions.
- Scenarios requiring rigorous self-reflection and result validation through automated evaluation loops.
- Building custom, scalable AI agent systems where individual components need to be upgraded independently.
How Orthogonal Agent Framework Works
- The Planning module decomposes a high-level user request into granular, actionable sub-tasks and sets clear goals.
- The Execution module invokes specific tools and operations to perform the planned steps sequentially.
- The Memory module manages both short-term context and long-term data storage to maintain state consistency.
- The Evaluation module reviews the execution results against predefined criteria and provides feedback for improvement.
- The system iterates through these modules, refining the plan or re-executing steps until the task is successfully completed.
Orthogonal Agent Framework Setup
To utilize this framework within Openclaw Skills, ensure your environment supports bash. You can initialize the framework by invoking the planning command for your specific task:
# Initialize a task with planning
正交性框架：规划 [Your Task Description]
# Execute a specific step
正交性框架：执行 [Step Number]
# Check current status of all modules
正交性框架：状态
Orthogonal Agent Framework Data Schema & Taxonomy
The framework maintains a transparent file-based data structure to track the agent's lifecycle. This is how the Openclaw Skills integration organizes its workspace:
| File | Description |
|---|---|
plan.md |
Contains task decomposition, goals, and current progress. |
execution.md |
A detailed log of tool calls, command outputs, and action results. |
memory.md |
Stores contextual information and long-term knowledge retrieved during the task. |
evaluation.md |
Stores scores, feedback, and validation results for each step. |
agent-state.yaml |
A technical manifest tracking active steps and the internal state of each module. |
Orthogonal Agent Framework Advanced Features
- Modular Extensibility: Easily add new dimensions like 'Creativity' or 'Research' to the existing workflow.
- Hot-Swappable Components: Switch from simple file-based memory to advanced vector databases without rewriting core logic.
- Standardized YAML Interfaces: Every module follows a strict interface (plan, execute, store, evaluate) for predictable behavior.
- Recursive Feedback Loops: The evaluation module can trigger the planning module to 're-plan' based on execution failures.
SKILL.md
Loading
Related Openclaw Skills
AI Trends Reporter
An automated report generator that synthesizes the latest AI news with personalized recommendations for high-value Openclaw Skills.
Tomorrow Weather & Almanac Push
An automated weather assistant that delivers daily Beijing forecasts, temperature fluctuation warnings, and traditional almanac insights via Feishu.
Meeting Score Skill
A professional meeting evaluation system that automates agenda scoring, judge management, and result summarization within Feishu Bitable.
AI PPT Annual Summary
An intelligent presentation generator that converts raw work logs into professionally designed annual summary PPTX files.
Restaurant Review Cross-Check
A powerful validation tool that cross-references Xiaohongshu and Dianping data to provide high-confidence restaurant recommendations.
SkillGate Governance
A robust supply-chain governance tool designed to scan, assess, and quarantine risky Openclaw Skills.
Featured*
