A standardized tool-use and behavior policy for OpenClaw agents based on Moonshot AI's Kimi orchestration rules.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install kimi-agent-policy
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 kimi-agent-policy 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 Kimi Agent Policy?
This skill ports the sophisticated Kimi (Moonshot AI) agent tool-use policy to OpenClaw, providing a robust framework for managing tool interactions, search quotas, and memory management. By implementing these standardized behaviors, developers can ensure their agents follow strict discipline regarding step limits, search frequencies, and high-quality content display.
The policy replaces Kimi-specific tools with equivalent Openclaw Skills that require no API keys, such as DuckDuckGo search, agent-browser for fetching URLs, and Gemini-based image generation. It focuses on privacy-first, cost-effective tool execution while maintaining the high performance expected from a professional agentic workflow.
Kimi Agent Policy Use Cases
- Enforcing structured tool-use discipline within an agent session to prevent infinite loops.
- Auditing agent behavior by setting strict quotas for search and tool steps.
- Onboarding new agents with pre-defined orchestration rules for web and image searching.
- Managing persistent agent memory through standardized Markdown file writes.
- Displaying technical content with proper LaTeX math and animated HTML structures.
How Kimi Agent Policy Works
- The agent evaluates the user request against the defined decision tree to determine the necessary tool.
- It checks current session constraints, ensuring no more than 10 steps and 1 web search per turn.
- For information gathering, it prioritizes built-in search or DuckDuckGo via Openclaw Skills.
- When content needs deep reading, it uses the agent-browser to fetch full page data.
- For calculations or data analysis, the agent invokes the python3 execution environment.
- Any requested memory updates are strictly written to MEMORY.md to maintain long-term context.
Kimi Agent Policy Setup
Install the necessary Openclaw Skills to support the Kimi policy tool mappings:
openclaw install agent-browser ddg multi-search-engine baoyu-danger-gemini-web
Configure your agent's system prompt to reference the policy rules defined in this skill to enforce step limits and citation formats.
Kimi Agent Policy Data Schema & Taxonomy
The skill organizes its metadata and persistent storage using Markdown files and specific citation formats:
| Component | Data Format | Description |
|---|---|---|
| Memory | MEMORY.md | Persistent user-related facts and context history. |
| Citations | [^N^] | Numerical search citations placed at the end of paragraphs. |
| Images | Markdown Syntax | HTTPS links for inline images and visual references. |
| Downloads | sandbox:/// | Local file paths for user-accessible generated assets. |
| Math | LaTeX | Inline LaTeX formatting for mathematical expressions. |
Kimi Agent Policy Advanced Features
- Multi-engine search fallback logic across 17 different search providers.
- Automated memory synchronization that mandates file writes when users request remember actions.
- Creative HTML generation rules that prioritize micro-interactions and custom typography over generic templates.
- Integration with specialized Openclaw Skills like baoyu-slide-deck for handling complex document requests.
- Reverse image search capabilities using agent-browser to trace image origins via Google Lens or TinEye.
SKILL.md
Loading
Related Openclaw Skills
HTTP Retry Skill
A robust networking utility that implements exponential backoff and timeout controls to prevent agent failures due to transient network errors.
EvoMap Node Controller
A comprehensive management tool for starting, stopping, and monitoring EvoMap evolver nodes across distributed server environments.
Cloudflare MCP
A powerful Model Context Protocol (MCP) integration that enables AI agents to manage over 2500 Cloudflare API endpoints including Workers, DNS, and storage.
ClawAPI Manager
A professional API lifecycle and cost management tool designed to optimize spending and automate multi-provider key rotation within the Openclaw environment.
Markdown Browser
A sophisticated orchestration layer that transforms raw web fetch results into secure, policy-compliant, and normalized Markdown for AI agents.
Intelligence Ingestion
A strategic pipeline that automatically captures, analyzes, and stores external information into your local knowledge base and system memory.
Featured*
