grounding-verification

Compare scrubbed versus full evidence to detect potential confabulation.

Updated Aug 27, 2026

Search Tags:#verification #audit #LLM #grounding #confidence #delta #claims

Ready to run

npx skills add https://github.com/gitwalter/cursor-agent-factory --skill grounding-verification

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: grounding-verification
Source: https://github.com/gitwalter/cursor-agent-factory/tree/main/.cursor/skills/grounding-verification
Command: npx skills add https://github.com/gitwalter/cursor-agent-factory --skill grounding-verification

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Universal two-pass grounding verification helps detect when an LLM's claims are unreliable by comparing a scrubbed, anonymized evidence pass with a full, detailed pass.

Core Features & Use Cases

Two-pass verification workflow: scrubbed vs full evidence, delta calculation, and a clear verdict (VERIFIED, PLAUSIBLE, SUSPICIOUS, or UNSUPPORTED).
Profile-driven thresholds for domains like code, documentation, data schemas, and security.
Deterministic, structured output suitable for integration and auditing.
Use Case: Validate software documentation claims against API schemas to guard against hallucinated documentation.

Quick Start

Prepare two passes of evidence (scrubbed/anonymized and full) for your LLM's claims, run the verification, and inspect the VERIFICATION REPORT for verdicts and Delta scores.

Frequently Asked Questions about grounding-verification

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

FAQPage Schema

How do I detect LLM hallucinations in generated documentation?▼

LLM hallucination detection compares scrubbed versus full evidence passes to flag confabulation. This grounding verification calculates a delta between the two passes to output a clear verdict, such as VERIFIED or UNSUPPORTED, for your claims.

What is two-pass grounding verification for LLM outputs?▼

Two-pass grounding verification is a deterministic method comparing an anonymized evidence pass against a full detailed pass. It calculates a delta score to determine if LLM claims are reliable and outputs a structured, auditable verdict.

How do I audit LLM claims against API schemas?▼

You audit LLM claims against API schemas by running them through a profile-driven verification process. Using domain-specific thresholds for code or data schemas, the process outputs a structured report indicating whether the claims are VERIFIED or SUSPICIOUS.

Does grounding verification work for both code and security domains?▼

Yes, grounding verification works across domains like code, documentation, data schemas, and security. It applies profile-based thresholds tailored to each specific domain to evaluate the reliability of LLM claims.

What do I need to provide to run a grounding verification check?▼

To run a grounding verification check, you need to provide frontmatter with a name and a verification profile. You must also prepare two passes of evidence, one scrubbed and one full, to compare against your LLM claims.