AI Content QA Gates: Grounding, Publication, Pinned Approval

    1

    And a draft that passed review can still change before it is published.

    $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

    AI Content QA Gates: Grounding, Publication, Pinned Approval

    Example session with this skill installed

    Check this generated product description against the supplier facts before it goes live: **Facts:** ["Plaque MDF Grow Where You Are Planted", "3/4 inch thick, 12 inches by 12 inches, Chocolate Edges"] Draft:

    A handmade square wall plaque with the words "Grow Where You Are Planted", a line from a 19th-century bishop. Made of solid MDF, 3/4 inch thick, 12 x 12 inches, with chocolate-brown edges. The perfect gift!

    • Read your context and instructions
    • Compiled the ai content qa

    Copy check: FAIL

    Grounding: 5 fault(s), 0 note(s)

    • FAULT claim: /\bperfect\b/i ("perfect")
    • FAULT claim: /!/ ("!")
    • FAULT claim: /\bhand[- ]?(made|crafted)\b/i ("handmade")
    • FAULT claim: /\bsolid\b/i ("solid")
    • FAULT number: 19

    Publication (html, policy content-qa-2026-10): pass

    • after sha256 3b1f...9c2e

    What the writer must do

    • Remove "handmade" and "solid": the facts do not say either. (If the supplier confirms handmade, add it to the facts; the gate then notes it instead of faulting it.)
    • Remove "19th-century": no number may appear that the facts do not have.
    • Drop "The perfect gift!": hype and exclamation marks are never allowed.

    Second pass, after the rewrite

    
    # Copy check: PASS
    ## Grounding: 0 fault(s), 0 note(s)
    ## Publication (html): pass
    

    Then through the store

    const draft = await store.add({ tenant: 'shop-a', targetType: 'product', targetId: 'gid://shopify/Product/4', field: 'descriptionHtml', oldValue: current, newValue: rewritten, publication: { kind: 'html' } });
    await store.approve(draft.id);   // publication gate runs; SHA-256 of newValue pinned
    await store.apply(draft.id);     // hash re-checked, gate re-run; now write to the platform with oldValue stored
    

    An edit to the draft file between approve and apply fails with "Content changed after approval; review again".

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

    About this skill

    A writer drafting from a few facts adds what sounds right: a size, a material, "since 1998", "handmade", "the best". On a live page a wrong claim is worse than a short one. And a draft that passed review can still change before it is published. This skill puts three gates between a draft and a page. They report; a person or a script decides. No model, no network.

    Grounding (grounded-copy.mjs)

    Every number in the copy must appear whole in the facts ("5" is not in "15"); prices, offers, free shipping, in stock, guarantees and hype are never allowed; handmade, certified, licensed, family-owned, organic, waterproof and the like are allowed only when the facts make the same claim, and then noted; HTML restricted to allowed tags with no attributes, balanced, no bare ampersand; word and length bounds; damaged characters; references the copy adds and watch-list words (materials, colours) noted for a reviewer. Every claim finding carries the words the copy used.

    Publication (publication-gate.mjs)

    Title 10 to 70, description 100 to 170, placeholders including upper-case TODO markers (Spanish "todo" is text), images without alt, javascript: and data:text/html resources, paragraphs over 90 words, packed list paragraphs, heading skips, JSON-LD that does not parse. A problem the field already had and the change leaves untouched is reported as inherited, not blocked, so legacy debt stays visible without stopping the fix.

    Review (draft-store.mjs)

    Draft, approve, apply, revert with the before value kept; approval runs the publication gate and pins the SHA-256 of the approved bytes; apply re-checks the hash and re-runs the gate, so an edit after approval goes back to review.

    CLI (check-copy.mjs)

    For CI or a pre-commit hook: copy file plus facts JSON in, Markdown or JSON report out, exit 1 blocks.

    Requirements

    Node 20+, no packages, no network. 11 offline tests. reference/gates.md explains each rule and how the three gates chain.

    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.

    English-only claim lists. Checks numbers and specific patterns, not all semantic fabrications. File-backed JSON store requires migration to SQLite for high-concurrency needs.

    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 19 days ago

    Software
    Node.js 20+

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    Shell runs node. Read files: the draft, the facts JSON and an optional before file you point it at. Write files: the JSON report and the draft store file. No network, no environment variables, nothing written to any platform.

    Listed19 days ago

    What's inside

    Frequently Asked Questions

    This skill uses deterministic Node.js code to verify AI outputs against your source data, catching the hallucinations and formatting errors that LLMs often miss when trying to self-correct.

    Because it is built with a zero-dependency architecture, it can be integrated into any Node 20+ environment, including CI/CD pipelines, custom LLM-ops workflows, and local agentic scripts.

    The integrity pinning feature generates a SHA-256 hash of your approved content; if even a single character is altered between the human approval step and the production deployment, the system will block the update.

    The package includes the core verification engine, a CLI for manual or automated audits, and a flexible configuration system for defining your specific publication standards.

    The grounding gate is currently optimized for English-language patterns and focuses specifically on validating numbers, dates, and defined claims rather than general semantic meaning.

    No, this tool operates entirely on your local infrastructure with no external model calls or network access required, ensuring your draft content and fact sheets remain private.

    New in LLM Ops