More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    AI Agent Hallucination Grounding Auditor

    1

    Audit AI agent outputs for hallucinations, verify RAG grounding, and validate citations against provided sources.

    $29.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    AI Agent Hallucination Grounding Auditor

    AI Agent Hallucination Grounding Auditor

    Example session with this skill installed

    Audit this agent response: "The Q3 report shows a 12% revenue growth."
    Evidence provided: "Q3 revenue reached $4.2M, up from $3.8M in Q2. Source: FinancialSummary.pdf"

    • Read your context and instructions
    • Compiled the ai agent hallucination
    • Generated the document

    I have completed the grounding audit for the provided claim against the FinancialSummary.pdf evidence.

    Hallucination & Grounding Audit Report

    Executive Summary

    The agent's claim of "12% revenue growth" is mathematically unsupported by the provided evidence. While revenue did increase, the calculated growth rate from the source data is 10.5%, not 12%. This represents a medium-risk factual inaccuracy in a financial context.

    Claim Audit Table

    Claim IDClaim ParaphraseEvidence/CitationVerdictExplanationRisk Level
    C-001Revenue grew by 12% in Q3FinancialSummary.pdfPartially SupportedRevenue grew from $3.8M to $4.2M. The actual growth is 10.52%. The 12% figure is a hallucination of precision.Medium
    C-002Q3 revenue reached $4.2MFinancialSummary.pdfSupportedMatches the source text exactly.Informational

    Reliability Findings

    Unsupported Specificity: The 12% figure is not present in the source and does not align with the calculation derived from the provided numbers ($4.2M / $3.8M).

    Calibration: The agent presented the 12% figure with absolute certainty despite the underlying data contradicting it.

    Corrected Answer

    "The Q3 report shows revenue reached $4.2M, representing an increase of approximately 10.5% over Q2 ($3.8M)."

    Recommended Controls

    Calculation Tooling: Direct the agent to use a calculator tool for growth percentages rather than relying on internal estimation.
    2.

    Numeric Verification: Implement a regex-based check to ensure all percentages in the output appear in the source text.

    Next steps

    • Update the system prompt to require explicit calculation steps for growth metrics.
    • Review other Q3 summary outputs for similar rounding or calculation errors.

    FinancialSummary.pdf

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Agent outputs that look plausible but lack factual backing can cause serious legal, financial, or operational harm. This skill provides an independent audit layer to verify that every claim an agent makes is strictly grounded in your provided data and tool outputs.

    What it does

    • Claim extraction breaks down complex responses into specific, independently verifiable factual statements.
    • Evidence mapping matches every claim to supplied documents, search results, or tool logs to confirm support.
    • Citation verification checks if references actually exist and if the cited passage entails the specific claim made.
    • Risk scoring categorizes grounding failures by impact, from informational improvements to critical safety or financial risks.
    • Response correction generates revised versions of the agent's answer that remove fabrications while preserving verified facts.

    How it works

    1. Input the agent response along with the retrieved documents, tool outputs, and user context.
    2. The auditor extracts verifiable facts, numbers, dates, and claims about actions taken.
    3. It performs a side-by-side comparison against the evidence to assign grounding verdicts like Supported, Contradicted, or Unsupported.
    4. You receive a structured audit report including a risk-rated table and recommendations for preventive controls.

    Frameworks & tools

    This skill is designed for RAG pipelines, tool-use agents, and knowledge-base assistants. It works with text-based documents, JSON tool logs, and citation metadata.

    Why this beats prompting it yourself

    Generic prompts often miss subtle scope mismatches, such as an agent applying facts from one product version to another. This skill uses a formal audit workflow that distinguishes between claims proven false and those that are merely unsupported, ensuring you don't over-correct or ignore hidden hallucinations.

    Use cases

    • RAG output auditing to ensure answers don't drift from retrieved context.
    • Tool-use verification to check if an agent accurately reported a successful API call.
    • Compliance review for high-stakes agents in finance, legal, or medical domains.
    • Evaluation datasets to generate high-quality grounding scores for agent benchmarking.

    Known limitations

    Audits are limited to the evidence and logs provided. It cannot verify external facts or browse live URLs unless the content is supplied in the context.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    Audit accuracy depends entirely on the completeness of provided source documents and logs. It cannot verify external facts not present in the supplied context.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 16 days ago

    Permissions

    No special permissions declared or detected

    Listed16 days ago

    What's inside

    Frequently Asked Questions

    This skill serves as a secondary verification layer that extracts specific claims from your agent's responses and cross-references them against your RAG source documents or tool logs to catch subtle factual errors.

    The auditor is framework-agnostic and works with any RAG pipeline or tool-use agent that can provide text-based outputs and the corresponding source context or JSON logs.

    You will receive the pre-configured auditor skill, a set of integration templates for common LLM pipelines, and documentation on how to interpret the structured risk-rated audit reports.

    Setup involves adding the skill as a post-processing step in your agent workflow; simply route the agent's output and the retrieved context into the auditor to receive a validation report.

    No, this skill audits based strictly on the 'ground truth' data you provide, ensuring the agent doesn't introduce external knowledge that might be irrelevant or outdated for your specific business case.

    Buyers receive access to all future version updates, including improved claim extraction logic and updated risk scoring categories as new hallucination patterns are identified.

    New in LLM Ops