dehallucination

Verify factual claims in artifacts and assign explicit confidence levels.

8|5|Updated Dec 30, 2025
One-click install
npx skills add https://github.com/axiomantic/spellbook --skill dehallucination
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: dehallucination
Source: https://github.com/axiomantic/spellbook/tree/main/skills/dehallucination
Command: npx skills add https://github.com/axiomantic/spellbook --skill dehallucination

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Dehallucination provides a structured verification framework to identify, assess, and correct false or unfounded claims within the Forge pipeline, ensuring outputs are grounded in evidence and sources.

Core Features & Use Cases

  • Factual verification: assigns explicit confidence levels to claims and requires citations or supporting context.
  • Hallucination detection: identifies fabricated references, invented capabilities, and false constraints.
  • Recovery & governance: prescribes explicit corrections, propagates lessons, and safeguards the Forge workflow.

Quick Start

Run the dehallucination workflow on a sample artifact to verify claims and generate a corrected version.

Frequently Asked Questions about dehallucination

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

FAQPage Schema
How do I verify factual claims and detect hallucinations in generated artifacts?

To verify factual claims and detect hallucinations, the workflow identifies fabricated references and assigns explicit confidence levels to each statement, producing a corrected artifact and a confidence map with supporting citations.

What is the best way to assign confidence levels to claims during a technology review?

Assigning confidence levels during a technology review requires capturing evidence and propagation paths for each claim, generating a verification report that flags false constraints and prescribes explicit recovery corrections.

How do I correct fabricated references and false constraints in my workflow outputs?

Correcting fabricated references involves running a hallucination detection process that identifies invented capabilities, applies explicit corrections, and propagates governance lessons to safeguard the workflow.

Does this hallucination detection process require any external dependencies or components?

This hallucination detection and recovery process operates with no external dependencies or components, structuring its verification framework directly within the pipeline to ground outputs in evidence.

When do I need to generate a confidence map for artifact verification?

You need to generate a confidence map for artifact verification when your technology review workflow requires explicit citations and recovery steps to prevent hallucinations and ensure outputs are grounded in sources.

What are the limitations of automated factual verification for preventing hallucinations?

Automated factual verification limitations include the need for explicit supporting context or citations for each claim, as the process relies on capturing evidence and propagation paths to identify false constraints and prescribe recovery steps.