A cognitive framework for AI agents to infer true user intentions and prevent misaligned or wasted work.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install intention-engine
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 intention-engine 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 Intention Engine?
The Intention Engine is a sophisticated reasoning layer designed for persistent AI agents to move beyond surface-level task execution. By distinguishing between the task (the surface) and the intention (the direction), this skill ensures that agents remain aligned with high-level user goals even when instructions are vague or technically incomplete. It allows your agent to bridge the gap between what is said and what is actually desired, making it a cornerstone of effective Openclaw Skills implementations.
This skill leverages context layering and premortem analysis to provide agents with a simulated gut feeling. Instead of blindly following instructions that might lead to failure or wasted resources, the Intention Engine prompts the agent to cross-reference multiple data sources—such as user profiles and recent memory—to validate the path forward. It transforms an agent from a simple command-executor into a strategic partner that understands the nuance of quality and the risk of negative intent.
Intention Engine Use Cases
- Resolving specification gaps where the user knows why they want something but is unclear on the how.
- Identifying intention gaps where a precise task is requested but the underlying purpose is unknown.
- Preventing the Klarna trap by checking for unstated constraints before optimizing for a specific metric.
- Managing long-running projects by re-validating intentions that have gone stale over a 30-day period.
- Strategic push-back when a requested task directly conflicts with the user's stated long-term profile goals.
How Intention Engine Works
- Classify the Gap: The agent first determines if the request lacks specification clarity, intention clarity, or both.
- Multi-Source Verification: It checks the user profile, active topic context, recent memory, and task state to infer the true intention.
- Mandatory Premortem: For any expensive or irreversible action, the agent identifies the most likely failure mode to mitigate risk.
- Quality Bar Assessment: The agent decides if the task requires a fast, adequate delivery or a highly polished, crafted output.
- Negative Intent Analysis: It evaluates what a bad version of success looks like to ensure unstated constraints are not violated.
- Redirection or Execution: If the task serves the intention, it executes; otherwise, it suggests a more efficient path or pushes back.
Intention Engine Setup
To integrate the Intention Engine into your agent workspace, follow these steps:
# Navigate to your agent's skill directory
cd ~/openclaw-agent/skills/
# Create the intention engine skill file
touch intention-engine.md
# Add the skill to your agent's active manifest
echo "- skills/intention-engine.md" >> agent-config.yaml
Ensure your agent has access to a USER.md file or equivalent user profile to maximize the effectiveness of this skill in your Openclaw Skills collection.
Intention Engine Data Schema & Taxonomy
The Intention Engine organizes its decision-making logic through the following data taxonomy:
| Component | Source | Function |
|---|---|---|
| User Goals | USER.md | Primary source for high-level alignment and priority checking. |
| Topic Context | Session Metadata | Identifies the current domain to narrow down inference possibilities. |
| Interaction History | Memory Log | Uses the last 2-3 days of decisions to maintain conversational momentum. |
| Freshness Flag | Internal State | Tracks intentions older than 30 days for mandatory re-validation. |
| Premortem Result | Runtime Analysis | A single-sentence evaluation of failure likelihood before execution. |
Intention Engine Advanced Features
- Context layering philosophy that requires at least two data sources before an intention is inferred.
- Automated quality bar distinction to prevent over-engineering of routine tasks.
- Negative intent checking to protect against optimizing for metrics while destroying unstated values.
- Intelligent push-back triggers that activate when tasks repeat previously failed patterns.
- Intention freshness monitoring to ensure the agent doesn't act on outdated objectives.
SKILL.md
Loading
Related Openclaw Skills
Activated Thinker
A sophisticated cognitive framework that equips AI agents with behavioral modes and advanced thinking protocols to solve complex problems more effectively.
GitNotesMemory
A silent, branch-aware knowledge graph memory system for AI agents that leverages git notes to persist context, decisions, and tasks across development sessions.
UI Control Center
A specialized local management interface for maintaining agent reliability, dashboards, and workflow monitoring.
Tradebot Tuning
A precision skill designed to safely optimize trading bot parameters and diagnostic gates to restore tradability in shifting market regimes.
Skill Detector
Skill Detector is an intelligent architect that passively monitors your workflows to auto-draft, audit, and improve your personal library of Openclaw Skills.
STM Assistant
A professional AI email assistant that manages outbound communication, branded signatures, and routing on behalf of a human user.
Featured*
