honest

Label AI-generated claims with confidence levels and trace their sources.

4|2|Updated Dec 31, 2025

Search Tags:#fact-checking #audit #hallucination #ai accuracy #confidence labels #epistemic discipline

Ready to run

npx skills add https://github.com/TylerGarlick/abraxas --skill honest

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: honest
Source: https://github.com/TylerGarlick/abraxas/tree/main/skills/honest
Command: npx skills add https://github.com/TylerGarlick/abraxas --skill honest

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill combats AI hallucination by making AI output transparent, labeling every claim with its confidence level and source, ensuring you can trust the information provided.

Core Features & Use Cases

Confidence Labeling: Every statement is tagged as [KNOWN], [INFERRED], [UNCERTAIN], or [UNKNOWN].
Source Tracing: Identifies the origin of claims to verify their grounding.
Session Auditing: Reviews entire conversations for unlabeled or fabricated claims.
Use Case: When researching a critical topic, use /audit to review the entire conversation and identify any claims that are not well-supported, ensuring your final conclusions are based on reliable information.

Quick Start

Use the honest skill to check the last response for confidence labels.

Frequently Asked Questions about honest

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I fact-check AI-generated text for hallucinations and accuracy?▼

Fact-checking AI-generated text involves labeling every claim with confidence levels like [KNOWN], [INFERRED], or [UNCERTAIN] and tracing its source origin. This process ensures transparency, allowing you to verify grounding and trust the output.

What is confidence labeling in AI accuracy and how does it work?▼

Confidence labeling tags every AI-generated statement as [KNOWN], [INFERRED], [UNCERTAIN], or [UNKNOWN]. This mechanism enforces epistemic discipline by making the reasoning boundaries transparent so you can assess claim reliability.

How do I audit a conversation for unsupported or fabricated AI claims?▼

Auditing a conversation for fabricated claims requires running a session-wide review command. This scans the entire text to identify unlabeled or unsupported statements, ensuring your conclusions rely on verified information.

Can I trace the source of AI claims to verify their grounding?▼

Yes, source tracing identifies the origin of AI claims to verify their factual grounding. By tracking where information comes from, you can distinguish between well-supported facts and potential hallucinations.

When do I need to apply epistemic discipline to AI outputs?▼

You need epistemic discipline when researching critical topics where accuracy is paramount. Applying confidence labels and context framing prevents hallucinations from undermining your final conclusions.

Does this fact-checking skill require any external dependencies to run?▼

No, this fact-checking skill operates without external dependencies. It relies entirely on its internal scripts and references to provide confidence labels, source tracing, and session-wide auditing.