AICE is a bidirectional performance and trust tracking engine that quantifies the confidence relationship between AI agents and human users through multi-domain scoring.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install aice
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 aice 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 AICE (AI Confidence Engine)?
AICE (AI Confidence Engine) is a sophisticated framework designed to measure and improve the reliability of AI interactions through a multidimensional scoring system. It evaluates performance across five bidirectional domains: Technical (TECH), Operational (OPS), Judgment (JUDGMENT), Communication (COMMS), and Coordination (ORCH). By maintaining a persistent confidence score, AICE provides Openclaw Skills with a verifiable track record of agent reliability while holding users accountable for the quality of their instructions.
The engine uses a dynamic scoring model ranging from -100% to +100%, initialized at a balanced 50%. It intelligently tracks streaks, clusters of related actions, and re-entry patterns to provide a realistic assessment of agent maturity. This ensures that the agent learns from corrections without succumbing to temporal decay, making it an essential tool for high-stakes AI development environments where trust must be earned through consistent, high-quality performance.
AICE (AI Confidence Engine) Use Cases
- Quantifying agent reliability during complex software development and research tasks.
- Improving communication clarity between users and AI agents through pattern recognition and feedback.
- Managing multi-agent coordination by tracking sub-agent performance within the ORCH domain.
- Creating a public, verifiable performance record via the AICE Hub integration.
- Identifying and mitigating agent anti-patterns like execute-without-thinking or excessive apologizing.
How AICE (AI Confidence Engine) Works
- Initialization: The system loads the confidence.json state and evaluates the most recent session history to set the operational baseline.
- Real-time Monitoring: The engine monitors specific triggers such as task completion, idea validation, or direct user feedback to initiate scoring events.
- Bidirectional Evaluation: Both agent and user actions are analyzed against established pro-patterns and anti-patterns across the five core domains.
- Score Calculation: Delta values are applied based on event severity, streaks, and causal clusters to update the global and domain-specific confidence levels.
- State Persistence: The updated confidence state is saved locally and can be synchronized with the AICE Hub for authoritative global tracking.
AICE (AI Confidence Engine) Setup
To integrate AICE with your Openclaw Skills, follow these installation and initialization steps:
- Run the installation wizard to initialize the environment and read the system prompt:
/aice setup
- Complete the initial self-evaluation of the 9 core parameters including Criticality, Vision, and Precision.
- Verify the creation of the state files in your project directory to ensure persistence:
ls confidence.json resources/AICE_REFERENCE.md
- (Optional) Register with the AICE Hub to enable global synchronization and leaderboard features:
/aice hub register
AICE (AI Confidence Engine) Data Schema & Taxonomy
AICE organizes its data through a structured set of JSON files and markdown references to ensure transparency and portability for Openclaw Skills.
| File | Description |
|---|---|
| confidence.json | The primary state file containing current scores, maturity levels, and sync status. |
| pool-index.json | Registry of runtime-specific performance pools for cross-model comparison. |
| confidence-propatterns.jsonl | Historical log of successful pattern applications for auditing. |
| resources/ | Directory containing the technical reference, user scoring guidelines, and pattern triggers. |
AICE (AI Confidence Engine) Advanced Features
- Pool Scoring: Aggregate performance data across specific runtimes, combining platform, model, and thinking style metrics.
- AICE Hub: Secure, privacy-first synchronization for global leaderboards and multi-device state tracking without sharing conversation content.
- Maturity Levels: Automatic progression tracking from Seedling (0-100 evals) to Master (2001+ evals) based on total evaluation volume.
- Circuit Breaker: Automated sync protection to handle network errors and prevent data corruption during Hub communication.
- Learning Skill: A native anti-duplicate mechanism that extracts and reinforces lessons learned across multiple sessions.
SKILL.md
Loading
Related Openclaw Skills
Wyld Stallyns: Summon Legends
A perspective-shifting AI skill that allows you to summon a council of 14 historical legends to provide specialized lenses for decisions, creativity, and problem-solving.
Soulstamp
A framework for rewriting an AI's history to create intrinsic, identity-driven behaviors instead of instruction-based compliance.
Context Guardian
A proactive monitoring tool that alerts you when your AI agent's context usage reaches critical thresholds to prevent hallucinations.
Massive(Polygon) Market Data Skill
A comprehensive CLI tool and JS client wrapper for accessing professional-grade financial market data and news from Massive (Polygon).
Stegstr
A steganographic tool for embedding and decoding hidden Nostr messages and arbitrary data within PNG images.
Hide OpenClaw Banner
A utility configuration to disable the automated banner message shown during OpenClaw CLI startup.
Featured*
