A sophisticated skill that allows AI agents to internalize your decision-making system and act autonomously based on your values and goals.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-direction
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 self-direction 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 Self-Direction?
Self-Direction is a core capability designed to bridge the gap between human intent and autonomous execution. By capturing your unique internal direction system—including values, risk tolerance, and resource priorities—this skill enables agents to move beyond simple rule-following. It transforms your agent into a collaborator that understands the why behind your requests, ensuring every action aligns with your long-term vision.
As part of the broader Openclaw Skills ecosystem, Self-Direction progressively builds a model of your decision-making patterns through a continuous learning loop. It observes your choices, validates its understanding, and applies these insights to future tasks, eventually allowing the agent to direct itself and even manage sub-agents with minimal manual intervention.
Self-Direction Use Cases
- When an agent needs to make high-level decisions without waiting for explicit user instructions.
- Scenarios where agents must understand the underlying rationale behind a task to ensure quality alignment.
- Maintaining consistent direction and behavioral standards across multiple sub-agents and complex workflows.
- Progressively automating resource allocation (time, tokens, and attention) based on historical user preferences.
How Self-Direction Works
- Observe: The agent monitors your decisions, corrections, and reasoning during natural interactions.
- Capture: It extracts patterns from these observations and records them into a structured direction system model.
- Validate: The agent checks its understanding by asking clarifying questions to refine its internal model.
- Apply: Once high confidence is reached, the agent uses the learned direction to execute future tasks autonomously.
- Transmit: The main agent generates transmission frames to pass relevant values and boundaries to sub-agents.
Self-Direction Setup
To get started with Self-Direction within the Openclaw Skills library, follow these steps:
- Initialize the skill by reviewing the internal setup.md file for specific integration guidelines.
- The system will automatically create the directory structure at
~/self-direction/upon first run. - Begin interacting with your agent; it will start capturing signals from your first session.
# Install via the hub
clawhub install self-direction
Self-Direction Data Schema & Taxonomy
The skill organizes its intelligence within the ~/self-direction/ directory using a transparent Markdown-based taxonomy:
| File | Description |
|---|---|
direction.md |
The core model containing values, goals, and criteria. |
evidence.md |
Raw logs of observations and interactions that informed the model. |
confidence.md |
Tracking of certainty levels (High, Medium, Low) for each captured element. |
conflicts.md |
A log of contradictory signals that require user resolution. |
transmission.md |
Summarized direction frames optimized for sub-agent consumption. |
Self-Direction Advanced Features
- Multi-Level Confidence Tracking: Automatically adjusts autonomy levels based on how certain the agent is about your preferences.
- Sub-Agent Direction Frames: Seamlessly propagates values and hard boundaries to spawned sub-agents to maintain alignment across teams.
- Context-Aware Inference: Distinguishes between different operational modes such as Learning, Autonomous, and Conservative.
- Proactive Gap Identification: The system identifies silences in your direction model and asks targeted questions to strengthen its decision-making.
- Conflict Resolution: Surfaces contradictory signals to the user to ensure the model evolves accurately over time.
SKILL.md
Loading
Related Openclaw Skills
Security Best Practices
A professional security review workflow that prioritizes exploitable risks and delivers minimal-diff fixes with verifiable evidence.
Second Order Effects
A strategic reasoning skill that identifies hidden risks and long-term opportunities by tracing decisions through three levels of consequences.
Seattle City Guide
A comprehensive intelligence layer for navigating Seattle as a tech professional, resident, or visitor, providing deep insights into neighborhoods, careers, and local culture.
Search Engine
A comprehensive framework for designing and building production-grade search engines with optimized indexing, retrieval logic, and relevance controls.
Self Discipline
A robust system for enforcing AI agent instruction compliance through root cause analysis and automated bash-based validators.
Self-Host
A comprehensive framework for deploying, securing, and maintaining resilient self-hosted services using containerization and automation.
Featured*
