Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Claim Checker: Fact-Check Text Against Its Sources

    2

    Fact-checks any text against its own links, with the source quote as evidence for every claim

    Free

    2 usesSecurity scanned
    Claim Checker: Fact-Check Text Against Its Sources

    Claim Checker: Fact-Check Text Against Its Sources

    Example session with this skill installed

    • Read your context and instructions
    • Compiled the claim checker

    Fact-check: Python's last two major releases

    6 claims checked: 3 verified, 3 partial.

    Quotes found on the downloaded source pages: 6 of 6.

    ⚠️ Partial: "Python 3.12 runs about 25% faster than Python 3.11"

    Source checked: https://www.python.org/downloads/release/python-3120/

    delivering an estimated 5% overall performance improvement

    Quote check: found on the page.
    Note: The release page says about 5%, not 25%.

    ⚠️ Partial: "distutils … had been deprecated since Python 3.8"

    Source checked: https://peps.python.org/pep-0632/

    In Python 3.10 and 3.11, distutils will be formally marked as deprecated.

    Quote check: found on the page.
    Note: The removal is right, but it was deprecated in Python 3.10, not 3.8.

    ⚠️ Partial: "Python 3.13 runs without the GIL by default"

    Source checked: https://docs.python.org/3/whatsnew/3.13.html

    This is an experimental feature and therefore is not enabled by default.

    Quote check: found on the page.
    Note: Free-threaded mode is experimental and off by default.

    ✅ Verified (3 claims)

    Release dates and the f-string change, each with its quote found on the page.

    Next steps: correct the three partial claims, or ask for suggested rewordings that the sources support.

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

    About this skill

    AI-written reports usually have a link after every claim, and they read well. But open the links and the source often says something slightly different: a different number, a different year, "experimental" instead of "default".

    claim-checker makes your agent open every cited page and mark each claim:

    • Verified: the page supports it, with the supporting quote
    • Partial: the source says something different (number, period, metric, or the claim goes too far)
    • Unverified: the page doesn't say it, or can't be opened

    It also lists figures that cite no source at all.

    Scripts keep it honest, not the model:

    • A script extracts every cited sentence, so no claim is skipped.
    • The report script downloads each cited page itself and refuses any quote it can't find there, so a made-up quote can't pass as verified.
    • The tally is counted by the script, not the model.
    • Every figure in a "verified" claim must also appear in its quote, so a real quote that states a different number ("25%" claimed, "5%" quoted) can't pass.
    • PDF sources (results releases, filings) are read too. If a page can't be downloaded (paywall, JavaScript-only page, or no web access for code), the report says so plainly instead of passing it.

    On the included example, a short piece about Python releases, it caught 3 of 6 claims misstating their sources, such as "25% faster" where the release page says "an estimated 5%".

    Free and open source (MIT), with tests.

    • It checks that each quote is on the source page and that a verified claim's figures appear in its quote; whether the quote fully supports the claim's wording is still the agent's judgement.
    • It can only verify claims that link to a source; uncited figures are listed, not checked.
    • Paywalled pages, pages that need JavaScript, and some sites that block automated downloads may show as "not machine-checked".
    • In the Claude app, the script's quote check needs code to be allowed on the web (Settings → Capabilities).

    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

    2 uses

    Times used to date

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 26 days ago

    • Free with an Agensi account

    Requires

    Python 3.10+

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access

    File Scopes

    *.md
    *.json

    Runs two small Python scripts. Network access is used to open the cited pages and to download them for the quote check. It only writes its own files (claims.json, check.md) in the working folder.

    Listed27 days ago

    What's inside

    Frequently Asked Questions

    No. It reports what each source says. You can ask for corrections afterwards.

    The claim is marked unverified or "not machine-checked", with the reason. It never passes silently.

    It lists figures with no source, but it can only verify claims that cite a page.

    New in Research & Analysis