AI.MD is a methodology and conversion tool that transforms human-readable system instructions into high-precision, structured labels optimized for LLM attention and token efficiency.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ai-md
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 ai-md 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 AI.MD: AI-Native Structured Instructions?
AI.MD is a specialized skill within the Openclaw Skills ecosystem designed to solve the paradox of decreasing AI compliance: as human-written instructions grow longer and more descriptive, AI performance often degrades. This skill applies a battle-tested conversion system that rewrites natural language prose into a structured-label format (e.g., triggers, actions, and exceptions).
By leveraging the way Large Language Models (LLMs) actually process information—through attention mechanisms rather than human-style reading—AI.MD ensures that every rule receives full attention weight. This approach has been proven across multiple models, including GPT-5.3 and Claude Opus 4.6, to increase Codex compliance from 6/8 to a perfect 8/8 while significantly reducing the token footprint of system prompts.
AI.MD: AI-Native Structured Instructions Use Cases
- Reducing high token consumption in long-running AI coding sessions.
- Fixing 'instruction drift' where the AI ignores specific rules in a large CLAUDE.md file.
- Standardizing system instructions across different LLM models (GPT, Gemini, Claude, Grok).
- Optimizing complex workflows involving multiple triggers, safety gates, and cross-references.
How AI.MD: AI-Native Structured Instructions Works
- The skill analyzes existing instructions as a state machine rather than a document, identifying atomic Triggers, Actions, and Constraints.
- It decomposes compound rules into single-concept lines to prevent 'Attention Splitting'.
- It assigns Zero-Inference Labels (like
trigger:,persist:, oryields-to:) to declare meaning explicitly without requiring model inference. - Rules are organized into a hierarchical structure using XML section tags like
<gates>,<rules>, and<rhythm>to create semantic boundaries. - The system handles hidden conflicts using a Priority Matrix to ensure safety gates always fire before courtesy actions.
- Finally, the converted output is validated against a multi-model exam protocol to ensure cross-platform reliability.
AI.MD: AI-Native Structured Instructions Setup
To integrate AI.MD into your Openclaw Skills library, follow these steps:
# Navigate to your project directory
cd your-project
# Run the distill command to analyze your current CLAUDE.md
openclaw run ai-md distill
# Or use the specific trigger phrases within your AI agent interface
# "distill my CLAUDE.md" or "rewrite my MD for AI"
AI.MD: AI-Native Structured Instructions Data Schema & Taxonomy
AI.MD organizes system instructions into an AI-native hierarchy for optimal parsing:
| Section | Function | Content Type |
|---|---|---|
<user> |
Persona & Identity | Key-Value pairs (e.g., tone: technical) |
<gates> |
Hard Constraints | Trigger-Action-Exception sequences |
<rules> |
Behavioral Guidelines | Atomic behavioral rules with function labels |
<rhythm> |
Workflow Patterns | Temporal constraints and recurring patterns |
<conn> |
Static Facts | Uncompressed credentials, URLs, or file paths |
<learn> |
System Evolution | Instructions for how the agent should adapt |
This schema ensures that Openclaw Skills maintain a 'Single Source of Truth' while using cross-references instead of rule duplication.
AI.MD: AI-Native Structured Instructions Advanced Features
- Bilingual Label Strategy: Uses English labels for universal LLM understanding while maintaining native-language output for users.
- State Machine Architecture: Treats instruction sets as logic gates with explicit
yields-topriority management. - Semantic Anchoring: Creates matchable tags that act like hash table lookups for faster, more accurate intent matching.
- Conflict Detection Matrix: Automatically identifies and resolves logical overlaps between different instruction sets.
- Token Burn Preview: Provides a real-time calculation of byte-to-token savings before and after conversion.
SKILL.md
Loading
Related Openclaw Skills
DingTalk Group Management Skill
A technical interface for managing DingTalk internal group chats through the DingTalk Open Platform API.
OpenClaw Immediate Ack
A specialized skill for OpenClaw agents that sends an immediate, bilingual acknowledgment message to users before the full AI response is generated.
Video Transcript
An automated tool for extracting, cleaning, and translating video subtitles into readable text transcripts.
image-gen: AI Image Generation & Editing
A comprehensive image generation and editing tool for AI agents that supports text-to-image, image-to-image, and automatic file optimization.
Romania
A specialized AI agent skill for planning deep, authentic Romanian travel experiences with regional expertise and practical logistics.
HTML Daily Report Generator Spec
A standardized HTML and CSS framework for AI agents to generate structured, masonry-style daily reports with built-in theme switching.
Featured*
