A machine-readable intent layer that ensures subagents optimize for your actual priorities rather than just completion metrics.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install intent-engineering
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 intent-engineering 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 Intent Engineering?
Intent Engineering provides a systematic framework to inject optimization priorities and tradeoff rules into Openclaw Skills. By defining an INTENT.md manifest, developers can prevent common AI failure modes where agents prioritize measurable metrics like response speed over qualitative goals like user value or brand loyalty. This skill ensures that every spawned subagent understands the core mission and knows exactly how to handle conflicting objectives during task execution.
At its core, this skill bridges the gap between high-level human goals and low-level agent execution. It adds a layer of intelligence that guides decision-making in multi-agent environments, ensuring that delegation, model selection, and quality standards remain consistent across an entire workspace using Openclaw Skills.
Intent Engineering Use Cases
- Setting up a new workspace to establish consistent subagent behavior and optimization rules.
- Preventing subagents from making incorrect tradeoffs between cost, quality, and speed during complex tasks.
- Implementing explicit control over model selection based on specific task categories.
- Managing complex delegation rules across multi-agent systems to ensure high-priority tasks are handled correctly.
- Aligning AI agent output with organizational values and long-term project goals.
How Intent Engineering Works
- The developer creates an INTENT.md manifest at the workspace root containing specific optimization priorities and tradeoff rules.
- The agent-context-loader library is integrated into the primary agent workflow within the Openclaw Skills environment.
- Every time a subagent is spawned, the prepareAgentContext function reads the current intent manifest fresh from disk.
- A compact, machine-readable intent summary (typically under 200 characters) is extracted and prepended to the subagent's task description.
- The subagent receives the instructions and executes the task while adhering to the defined priorities, such as prioritizing honesty over cost saving.
Intent Engineering Setup
Step 1: Create INTENT.md
Copy the intent template to your workspace root:
cp $(dirname $0)/references/intent-template.md $OPENCLAW_WORKSPACE/INTENT.md
Step 2: Install agent-context-loader
Copy the reference implementation to your library directory:
cp $(dirname $0)/references/agent-context-loader-template.js $OPENCLAW_WORKSPACE/lib/agent-context-loader.js
Step 3: Wire into spawn calls
Integrate the context loader into your subagent spawn logic using JavaScript:
const { prepareAgentContext } = require('./lib/agent-context-loader');
const { context } = prepareAgentContext(taskType, workspaceRoot);
const fullTask = context ? context + '\n\n---\n\n' + originalTask : originalTask;
// Execute subagent with fullTask
Intent Engineering Data Schema & Taxonomy
The skill utilizes a structured manifest file to govern Openclaw Skills behavior. The INTENT.md file organizes data as follows:
| Field | Purpose |
|---|---|
| optimization_priority | Defines an ordered list of priorities (e.g., user_value > honesty > cost). |
| tradeoffs | Explicit rules for when objectives conflict, such as speed vs. depth. |
| model_tier_intent | Specifies which model tiers should be used for different task types. |
| delegation_intent | Rules for when to delegate tasks to subagents versus handling them inline. |
| quality_intent | Domain-specific quality standards for architecture, coding, or research. |
Intent Engineering Advanced Features
- Silent degradation: If the INTENT.md file is missing, the system automatically injects safe defaults to maintain workflow continuity.
- Dynamic intent propagation: Summaries are generated in real-time on every call, allowing for flexible adjustments to agent behavior.
- Context-aware loading: Automatically extracts and summarizes only the relevant portions of the intent manifest based on the current task type.
- Multi-agent orchestration: Designed to work seamlessly across complex Openclaw Skills implementations involving multiple layers of subagent delegation.
SKILL.md
Loading
Related Openclaw Skills
Correction Memory
A persistent memory system that logs agent corrections and automatically injects them into future sessions to prevent recurring errors.
Agent Stability Framework (ASF)
A comprehensive three-layer defense system to prevent AI personality drift, technical faults, and soul misalignment in autonomous agents.
Video Watcher
A specialized tool for programmatically fetching and reading transcripts from YouTube and Bilibili videos to enable AI-powered analysis.
Joan Workflow
A workspace-based knowledge and task management system designed specifically for AI-assisted development workflows.
Cost Governor
A pre-flight cost estimation and approval gate system designed to prevent API overspend during multi-agent deployments.
Skill Vetter
A security-first vetting protocol designed to audit AI agent skills for malicious code, credential theft, and unauthorized data exfiltration before installation.
Featured*
