Cited-Answer Source Freshness Check for RAG

    1

    A retrieval-backed assistant ingests a page, answers from the snapshot, and cites the page.

    $29

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Cited-Answer Source Freshness Check for RAG

    Example session with this skill installed

    Check the freshness of the pricing-data pack in the ./packs folder and output the results as JSON.

    • Read your context and instructions
    • Compiled the cited-answer source freshness

    I have completed the freshness check for the requested pack. The scan identified one critical figure discrepancy where a price in the stored text no longer appears on the live source page.

    Freshness Report

    {
      "checkedAt": "2023-10-24T14:30:00.000Z",
      "findings": [
        {
          "pack": "pricing-data",
          "url": "https://example.com/fees/pro-plan",
          "level": "figures",
          "detail": "Stored figure '$49.99' not found in live text or script bundles."
        },
        {
          "pack": "pricing-data",
          "url": "https://example.com/docs/api-v1-legacy",
          "level": "gone",
          "detail": "Page returned status 404."
        },
        {
          "pack": "pricing-data",
          "url": "https://example.com/dynamic-dashboard",
          "level": "unverifiable",
          "detail": "Live page has fewer than 40% of stored words; likely requires JS rendering."
        }
      ]
    }
    

    Next steps

    • Update the pro-plan record in your vector database with the new pricing found on the live site.
    • Remove or redirect the citation for the 404 api-v1-legacy page.
    • Schedule a manual crawl or use a headless browser for the dynamic-dashboard URL to verify content.

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

    About this skill

    A retrieval-backed assistant ingests a page, answers from the snapshot, and cites the page. When the page changes (a fee, a closing time, a deadline), every later answer is confident, sourced and wrong, and nothing in the pipeline notices: the index is intact, the citation resolves, the user trusts it.

    Checker (freshness.mjs): for each ingested URL, fetches the live page and up to four of its same-origin script files (an i18n bundle is just strings and is where many sites keep their hours and dates), strips scripts, styles, navigation, header and footer, compares the stored answer text with the live words, and extracts figures (numbers, times, percentages, money) from both. Findings are tuned to be believed: figures when a figure you still answer with is gone from the page (the one worth waking someone for), gone for non-2xx, unverifiable when the page builds its text in the browser (too few live words, or more than 60% of the figures missing at once, which is a fetch that cannot see the page, not every fee changing), unreachable for network errors, and drift for wording differences only with --verbose, because curated prose is expected to differ. Han text is compared as 2-character grams so Chinese pages do not collapse into one word.

    CLI (freshness-cli.mjs): --packs dir or --pack file, --only, --json, --verbose; findings de-duplicated per pack, URL and level; exit 1 only for actionable findings.

    5 tests: tokenization incl. CJK, figure extraction, every comparison outcome (same, one changed fee, hours living in a bundle, browser-rendered page, all figures gone, reworded prose with and without verbose), a loopback HTTP server with same, changed, bundled, missing and unreachable pages, and the CLI's exit codes. Pure comparison functions are exported for your own tests; fetch is injectable.

    Node 18+, no packages. Network only to the URLs in your packs.

    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.

    Requires Node 18+. No JS rendering; client-side apps are marked unverifiable. Only compares text presence, not semantic meaning. Limited to 4 same-origin scripts per page.

    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 17 days ago

    Software
    Node.js 18+

    Permissions

    Network Access
    Read Files
    Write Files
    Listed17 days ago

    What's inside

    Frequently Asked Questions

    This tool is a zero-dependency Node.js utility designed for programmatic use. It integrates into any RAG pipeline via a simple CLI command or by importing the `checkArticles` function into your existing Node.js workflows.

    Unlike LLM-based verification, this tool uses deterministic local comparison and bigram matching to identify missing figures and text drift. This eliminates token costs, reduces latency, and prevents the 'hallucinated updates' that often occur when asking an AI to verify its own sources.

    The tool includes specific support for Han characters using bigram comparison, making it highly effective for Chinese-language knowledge bases. It also features a script inspector that can find data hidden in JS bundles, which standard scrapers often miss.

    The purchase includes the complete Node.js source code (including the CLI and core logic), documentation on JSON schema requirements, and a set of example scripts to help you integrate the tool into your nightly validation pipeline.

    The tool requires Node.js 18 or later and uses standard fetch and native test runners. It is designed to be lightweight with zero external dependencies, ensuring it won't bloat your project's dependency tree or cause version conflicts.

    New in LLM Ops