What does the confidence percentage represent?

The percentage represents a calculated score based on response completeness, language certainty, error detection, and session coherence.

Can I disable Inference Confidenz temporarily?

Yes, you can disable the skill by setting the CONFIDENZ_ENABLED environment variable to false or updating your configuration file.

What is the difference between minimal and detailed display?

Minimal display shows a brief code and percentage (e.g., CZ 87%), while detailed display includes the qualitative level like [high] or [low].

How do I customize the confidence thresholds?

You can define custom high and medium thresholds by creating or editing the .claude/confidenz.config.json file in your project root.

Inference Confidenz

Name: Inference Confidenz
Author: BLERBZ

byBLERBZ

Analíticas y Monitorización

Monitors and displays quantitative confidence levels for Claude Code responses to ensure output reliability.

Acerca de

Inference Confidenz is a specialized monitoring tool designed to provide transparency into Claude's internal certainty when generating code or documentation. By analyzing response quality, task completeness, and the presence of hedging versus definitive language, it assigns a confidence percentage (0-99%) to every interaction. This allows developers to instantly gauge the reliability of an AI-generated solution, making it easier to identify potential hallucinations or areas requiring manual review during complex development workflows.

Características Principales

0 GitHub stars
Automatic detection of error indicators and unresolved issues
Support for minimal and detailed display formats in the terminal
Configurable thresholds for high, medium, and low confidence levels
Detailed analysis of hedging vs. definitive language patterns
Real-time confidence scoring (0-99%) for every response

Casos de Uso

Monitoring context coherence across long-running development sessions
Identifying potential AI hallucinations through low-confidence alerts
Validating complex code refactors where high-certainty output is critical

Acerca de

Características Principales

0 GitHub stars
Automatic detection of error indicators and unresolved issues
Support for minimal and detailed display formats in the terminal
Configurable thresholds for high, medium, and low confidence levels
Detailed analysis of hedging vs. definitive language patterns
Real-time confidence scoring (0-99%) for every response

Casos de Uso

Monitoring context coherence across long-running development sessions
Identifying potential AI hallucinations through low-confidence alerts
Validating complex code refactors where high-certainty output is critical