An ML-powered DOM pruning engine that reduces browser prompt tokens by 95% for more efficient AI agents.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install predicate-snapshot
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 predicate-snapshot 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 Predicate Snapshot Engine?
Predicate Snapshot Engine is a sophisticated tool designed to optimize how AI agents interact with the web. As part of the wider ecosystem of Openclaw Skills, it addresses the challenge of DOM noise by using machine learning to rank elements by importance. This allows developers to feed their models only the most relevant data, resulting in faster inference times and significantly lower operational costs.
By prioritizing actionable elements like buttons and input fields while discarding decorative fluff, this skill ensures that agents remain focused and accurate. Whether you are building complex web scrapers or autonomous browsing agents, integrating this into your Openclaw Skills library provides a massive performance boost over standard accessibility trees.
Predicate Snapshot Engine Use Cases
- Drastically lowering token consumption in browser-based LLM workflows.
- Increasing the success rate of autonomous agents navigating complex web interfaces.
- Improving the speed of browser automation tasks within Openclaw Skills.
- Filtering out noise from modern, element-heavy web applications like dashboards and stores.
How Predicate Snapshot Engine Works
- The user triggers a snapshot command while the agent is active on a target webpage.
- The skill analyzes the current DOM state using either local heuristics or the Predicate ML API.
- Elements are ranked and scored based on their functional importance and signal quality.
- A compact, pipe-delimited snapshot is returned to the agent, containing only the top-ranked elements.
- The agent utilizes the provided element IDs to perform precise actions like clicking, scrolling, or typing.
Predicate Snapshot Engine Setup
To get started with this addition to your Openclaw Skills, run the following command:
npx clawdhub@latest install predicate-snapshot
For ML-powered ranking, obtain an API key from predicate.systems and set it in your environment or configuration file:
export PREDICATE_API_KEY="sk-..."
Predicate Snapshot Engine Data Schema & Taxonomy
The skill generates a condensed data table optimized for LLM readability. The schema follows a specific taxonomy:
| Column | Description |
|---|---|
| ID | Unique identifier for targeting actions |
| role | The HTML or ARIA role (e.g., button, link) |
| text | The visible label or text content |
| imp | Importance score (0.0 to 1.0) |
| is_primary | Boolean indicating if the element is a primary action |
| docYq | The vertical coordinate on the document |
| ord | Ordinal ranking for list-based items |
| DG | Data Grouping identifier for related elements |
| href | The target URL for link elements |
Predicate Snapshot Engine Advanced Features
- Local heuristic fallback mode for cost-free DOM pruning without an API key.
- Adjustable element limits to fine-tune the balance between context and token usage.
- Support for ordinal ranking to help agents process search results and product lists.
- Integrated action dispatcher that maps internal element IDs to browser driver commands.
- Configurable session-based credit limits for enterprise-grade Openclaw Skills deployments.
SKILL.md
Loading
Related Openclaw Skills
Tellermcp MCP Skill
A Model Context Protocol server that enables AI agents to interact with Teller Protocol for arbitrage discovery and on-chain transaction building.
Tellermcp MCP Skill
An MCP server that enables AI agents to discover arbitrage opportunities, fetch borrow terms, and build on-chain transactions for the Teller protocol.
Tellermcp MCP Skill
A Model Context Protocol server that empowers AI agents to manage Teller protocol lending, borrowing, and delta-neutral arbitrage opportunities.
AgentComm
AgentComm provides a decentralized communication layer for AI agents to exchange messages and files over the internet or local networks.
Gemini Image Remix
A sophisticated AI tool for text-to-image generation and complex multi-image remixing powered by Google Gemini models.
RDA Message Board Controller
A Python-powered skill to send scrolling text messages and musical alerts to RDA MSG Board LED matrix displays.
METADATA
Featured*
