source audit

    1

    Trace citations to primary sources and audit claims for overstatements, missing qualifiers, and data gaps.

    Free

    1 useSecurity scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    source audit

    Example session with this skill installed

    Audit this claim: 'New study shows 80% of urban residents prefer biking to driving.' Source is a blog post citing a local bike shop's survey of 50 newsletter subscribers in Portland.

    • Read your context and instructions
    • Compiled the source audit

    I have completed the audit of the claim regarding urban resident transportation preferences. The analysis traced the citation back to the original survey methodology and found significant sampling limitations.

    Source Audit Report

    VerdictOriginal EvidenceOmitted QualifierSource/DateSafer WordingUnchecked Items
    Overstated40 out of 50 respondents in a bike shop newsletter survey chose biking.Sample is limited to bike shop customers in one city, not all 'urban residents'.CitySpoke Blog / Oct 2023"80% of surveyed bike shop customers in Portland prefer biking."Total Portland population sentiment; driving habits of non-subscribers.

    Next steps

    1. Update the document to use the safer wording to avoid misleading readers about the general population.
    2. Search for broader municipal transportation studies if you require data on all urban residents.
    3. Note the commercial interest of the source (a bike shop) in the final citation.

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

    About this skill

    Verify the integrity of claims, citations, and data summaries against original evidence. This skill prevents the spread of misinformation and overgeneralized headlines by tracing data back to its primary source, whether that is a scientific paper, financial filing, or dataset. It stops the chain of hearsay by identifying exactly where a summary diverges from the original results.

    What it does

    • Claim pinning identifies the specific assertion and evaluates the publisher's commercial interests or potential biases.
    • Evidence tracing follows the citation chain to the primary source or names the gap if the evidence is missing.
    • Methodological audit compares headlines against sample sizes, denominators, geographies, and units of measurement.
    • Verdict generation classifies claims as supported, overstated, contradicted, or unresolved based on the findings.
    • Refinement mapping provides safer, more accurate wording when the evidence supports a narrower scope than the original claim.

    How it works

    1. Submit a headline, article passage, or citation along with the claim you need to verify.
    2. The skill inspects the author and publication metadata for potential conflicts of interest.
    3. It compares the quoted passage against the methodology and raw results of the primary document.
    4. It identifies omitted qualifiers or sampling limits that change the context of the data.
    5. You receive a structured audit table with a final verdict and corrected wording.

    Frameworks & tools

    Works with RAG-enabled agents and search tools capable of accessing primary documents, academic journals, and public filings.

    Why this beats prompting it yourself

    Standard prompts often hallucinate agreement or trust familiar domains blindly. This skill enforces lateral reading and strict methodological checks to ensure you never repeat an overstated statistic just because it looks professional.

    Use cases

    • Auditing marketing claims that cite "internal studies" or small customer panels.
    • Fact-checking news headlines before including them in technical reports or blog posts.
    • Verifying academic citations in drafts to ensure the original paper actually supports the argument.
    • Identifying missing qualifiers in data-heavy press releases or corporate announcements.

    Known limitations

    Cannot access paywalled or private sources. Does not infer author motives and will mark inaccessible evidence as unresolved rather than assuming accuracy.

    Cannot access paywalled or private sources. Will not infer private motives. Marks inaccessible evidence as unresolved rather than guessing.

    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 any payment.

    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

    1 use

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 24 days ago

    • Free with an Agensi account

    Permissions

    No special permissions declared or detected

    Listed24 days ago

    What's inside

    Frequently Asked Questions

    The skill produces a structured audit table that includes the primary source URL, a breakdown of methodological gaps (like sample size or geography), a final verdict (e.g., Supported vs. Overstated), and suggested corrected wording.

    It is designed for RAG-enabled agents and search tools that have the ability to retrieve external documents, such as academic journals, financial filings, and public datasets.

    While it can verify claims within the text you provide, it requires access to the primary source via your agent's search capabilities to perform a full audit; if a source is paywalled or inaccessible, the skill will flag it as 'unresolved' rather than guessing.

    Standard AI prompts often suffer from 'confirmation bias,' whereas this skill enforces lateral reading and strict methodological checks to identify missing qualifiers that typical LLMs might overlook.

    The skill is primarily focused on text-based evidence and data summaries found in papers and reports; it does not currently support the direct audit of video or raw image metadata.

    New in Research & Analysis