A high-discipline execution framework that balances agent autonomy with human trust through a rigorous 8-step control loop and preference learning.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ibt
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 ibt 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 IBT: Instinct + Behavior + Trust?
IBT (Instinct + Behavior + Trust) is a sophisticated execution framework designed for AI agents that require both technical discipline and nuanced judgment. It moves beyond simple prompt-response cycles by implementing a comprehensive control loop: Observe, Parse, Plan, Commit, Act, Verify, Update, and Stop. This structure ensures agents act with initiative without becoming reckless, especially in multi-step or high-impact environments. Integrating Openclaw Skills like IBT allows developers to build more reliable and predictable autonomous systems.
The framework's latest version introduces Preference Learning, which captures both explicit and implicit user preferences to reduce repetitive clarifications. By storing these preferences in a human-readable USER.md file, IBT ensures that agent behavior aligns with user expectations while maintaining strict privacy and security boundaries. It provides a blueprint for creating agents that understand when to push forward and when to wait for human approval.
IBT: Instinct + Behavior + Trust Use Cases
- Managing multi-step, trust-sensitive coding or deployment tasks where precision is required.
- Automating high-impact work that necessitates approval gates and destructive action previews.
- Reducing communication friction by learning user-specific response tones, lengths, and formats.
- Handling complex error recovery where transient failures need retries but permanent blockers need human intervention.
- Maintaining context continuity across long sessions using natural realignment protocols.
How IBT: Instinct + Behavior + Trust Works
- Observe: The agent assesses the request to notice risks, form a stance, or suggest alternative approaches before starting.
- Parse: The agent identifies the real goal and success criteria, prioritizing the resolution of goal-critical ambiguity.
- Plan & Commit: The agent selects the shortest verifiable path and clearly states its intention before proceeding.
- Act & Verify: The agent executes the plan cleanly and checks results against hard evidence rather than assumptions.
- Update & Stop: The agent patches the smallest failed steps if necessary and halts once success criteria are met or a blocker is reached.
IBT: Instinct + Behavior + Trust Setup
To integrate this skill into your agent environment, use the following CLI command:
clawhub install ibt
Once installed, the agent will reference the provided markdown files to govern its behavior and begin utilizing the USER.md file for preference storage.
IBT: Instinct + Behavior + Trust Data Schema & Taxonomy
The IBT framework organizes its data and operational doctrine across several specific files to ensure transparency and human control:
| File | Purpose |
|---|---|
| SKILL.md | The full framework definition, including core loops and safety rules. |
| USER.md | Stores learned preferences including communication style and project context. |
| POLICY.md | A concise, high-level operational doctrine for the agent. |
| TEMPLATE.md | A drop-in template for implementing IBT policies in new environments. |
| EXAMPLES.md | Practical examples of IBT behavior in various scenarios. |
IBT: Instinct + Behavior + Trust Advanced Features
- Preference Learning: Automatically captures patterns in user feedback to adjust response length, tone, and formatting.
- Trust Calibration: Dynamically adjusts agent autonomy and explanation depth based on task risk and evidence levels.
- Error Resilience: Classifies failures as temporary, permanent, or trust-related to determine the most logical response.
- Discrepancy Reasoning: A structured process for resolving data mismatches between agent tools and user-provided information.
- Approval Gates: Mandatory wait points for destructive, external, or high-cost actions to ensure human oversight.
- Trust Recovery: A non-defensive protocol for acknowledging mistakes and proposing safe corrections.
SKILL.md
Loading
Related Openclaw Skills
ARTA — Agentic Real-Time Awareness
ARTA provides an in-memory awareness layer for AI agents to track and query their own activities across multiple sessions within a single process.
Azure DevOps
A powerful integration for managing Azure DevOps projects, repositories, and pull request lifecycles directly through an AI-driven interface.
Xiaohongshu MCP
A Model Context Protocol (MCP) service for programmatically managing Xiaohongshu accounts, content publishing, and user engagement.
Qwen TTS (Local AI Voice Synthesis)
A powerful local text-to-speech engine powered by Qwen3-TTS for generating emotional and stylistically controlled audio.
Space Login
A comprehensive simulation framework for managing astronaut authentication and orbital mission telemetry across the solar system.
Space Login
A comprehensive simulation skill for managing astronaut authentication and orbital synchronization across the solar system.
METADATA
Featured*
