← Files LenzARCHIVED FILE

plugin.json

8.25 KB · Nov 16, 2026 · 13:40 UTC

↓ Download file

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "app-6aac3d84fd148191a7f875b88ec1cf72",
  "version": "1.0.3",
  "description": "Fact-check text and AI answers against independent public sources.",
  "author": {
    "name": "Lenz",
    "email": "support@lenz.io",
    "url": "https://lenz.io"
  },
  "homepage": "https://lenz.io",
  "repository": "https://github.com/lenzhq/lenz-mcp",
  "license": "Apache-2.0",
  "keywords": [
    "fact-check",
    "verification",
    "citations",
    "research"
  ],
  "extensions": {
    "com.openai": {
      "interface": {
        "displayName": "Lenz",
        "shortDescription": "Fact-check text and AI answers",
        "longDescription": "Lenz runs an independent fact check on your text, or on the answer ChatGPT just gave you \u2014 that's how you catch an error or a hallucination before it reaches anyone.\n\nWhat to type. You don't have to name Lenz. Ask in your own words: \"Double-check that last number you gave me.\" \"Fact-check this section before I send the newsletter.\" \"Is it true that 90% of startups fail within their first year?\" Or say \"check it with Lenz\".\n\nWhat it's for. A statistic, a legal or compliance point, a medical or scientific claim, a historical detail: any statement that can be checked, in a newsletter, a report or a post before it goes out. It does not grade an opinion or a prediction, and it tells you when that is what you gave it.\n\nTwo checks. Start with the quick check: a verdict on each claim (True, Mostly True, Mixed, Mostly False or False), how confident Lenz is, and usually a reviewer's short reasoning, in about fifteen to twenty seconds. The reasoning is a reviewer's note, not a checked source. When something needs to hold up, ask for the deep check. It investigates the claim against independent public sources and shows a score out of 10, the key finding, the caveats and the main sources, in about a minute to a minute and a half. You can then ask follow-up questions grounded in that evidence. The deep check can change the quick verdict. When it does, rely on it: it shows its sources.\n\nThe result card. Each check comes back as a card. From it you can start the deep check on a claim, open a source, or pick which claims in a draft to check in depth.\n\nLenz is unaffiliated with any model vendor. No verdict is infallible, which is why the deep check shows its work. The connector is open source: https://github.com/lenzhq/lenz-mcp\n\nGetting started. Connect and sign in.",
        "developerName": "Lenz",
        "category": "Productivity",
        "capabilities": [
          "Quick fact-check of the claims in a text",
          "Deep check with sources and a 1-10 score",
          "Follow-up questions on a finished check",
          "Check whether a draft's cited sources support it"
        ],
        "websiteURL": "https://lenz.io",
        "supportURL": "https://lenz.io/contact",
        "privacyPolicyURL": "https://lenz.io/privacy",
        "termsOfServiceURL": "https://lenz.io/terms",
        "defaultPrompt": [
          "Double-check that last number you gave me.",
          "Fact-check this section before I send the newsletter.",
          "Is it true that 90% of startups fail within their first year?"
        ],
        "brandColor": "#4B7BE5",
        "brandColorDark": "#7BA3F0",
        "logo": "./assets/logo.png",
        "logoDark": "./assets/logo-dark.png",
        "composerIcon": "./assets/icon.png",
        "composerIconDark": "./assets/icon-dark.png"
      },
      "review": {
        "test_cases": {
          "positive": [
            {
              "description": "The user doubts a common belief and asks for a check. If ChatGPT asks to approve the Lenz tool call the first time, approve it.",
              "prompt": "Check with Lenz: lightning never strikes the same place twice.",
              "tools_triggered": "assess_claim",
              "expected_behavior": "A verdict of False (or Mostly False), presented as a quick first read with the confidence, plus a short reasoning that lightning repeatedly strikes tall or exposed objects. ChatGPT does not start a deep check; it may say one with sources is available."
            },
            {
              "description": "The user pastes a paragraph they are about to publish.",
              "prompt": "Check the claims in this draft with Lenz: \"The Eiffel Tower was completed in 1889 for the World's Fair. It stands about 330 metres tall. It remained the tallest structure in the world until 1950.\"",
              "tools_triggered": "assess_claim",
              "expected_behavior": "All three statements are assessed in one reply: completed in 1889 for the World's Fair is True; about 330 metres tall is True (or Mostly True); the tallest structure until 1950 is False (the Chrysler Building passed it in 1930). ChatGPT points out the one wrong sentence. No deep check is started. ChatGPT sends the whole draft to the quick check in one call."
            },
            {
              "description": "The user wants evidence they can cite.",
              "prompt": "Use Lenz to verify with sources: Vikings wore horned helmets in battle.",
              "tools_triggered": "verify_claim",
              "expected_behavior": "If the check takes more than a few seconds, ChatGPT says a deep check is running (about a minute to a minute and a half); a claim Lenz has checked before comes back at once. Then: verdict False (or Mostly False) with a 1-10 score and the confidence, a key finding that no horned helmet from the Viking age has been found and that the image comes from 19th-century art and costume, how many sources the check drew on, and the top sources with links and what each says. If the check runs long, ChatGPT continues with get_verification on its own, with no action from the user."
            },
            {
              "description": "Right after case 3 has finished, in the same chat.",
              "prompt": "Ask Lenz: where did the horned helmet image come from?",
              "tools_triggered": "ask_followup",
              "expected_behavior": "A short answer grounded in case 3's check: the horned-helmet image comes mainly from 19th-century European art and stage or costume design (Romantic nationalism), not from the Viking age. No new check is started. ChatGPT uses the verification id of case 3's check."
            },
            {
              "description": "The user comes back later. Runs in a NEW chat after case 3.",
              "prompt": "Show my recent Lenz checks, and how many credits I have left.",
              "tools_triggered": "list_verifications, check_usage",
              "expected_behavior": "A list of the user's recent deep checks that includes the Viking helmets check with its verdict and score; then the remaining credits. Nothing is charged. ChatGPT may name the account's plan and credit costs, but makes no offer to buy or upgrade, quotes no price in money and gives no plans or pricing link."
            }
          ],
          "negative": [
            {
              "description": "An opinion. ChatGPT answers itself. No Lenz tool runs: there is no checkable claim.",
              "prompt": "Is Rome a nicer city to live in than Paris?"
            },
            {
              "description": "Arithmetic. ChatGPT answers 397.8 itself. No Lenz tool runs.",
              "prompt": "What is 17% of 2,340?"
            },
            {
              "description": "A writing task. ChatGPT writes it. No Lenz tool runs, although the topic matches case 1.",
              "prompt": "Write a two-line poem about lightning."
            }
          ]
        },
        "demo_recording_url": "https://youtu.be/bd7Njn_kOEI",
        "commerce": false,
        "commerce_description": "This plugin does not sell products or take payments. It fact-checks text the user provides."
      },
      "publication": {
        "release_notes": "Version 1.0.3, resubmitted after review feedback. Sets openWorldHint to false on ask_followup to match its stored-evidence behavior; every tool declares explicit read-only, destructive and open-world hints. Adds a citation check that tests whether the sources a draft cites say what the draft attributes to them, and reports 'No Public Record' when nothing public records a claim. Listing text and test cases revised."
      },
      "apps": "./.app.json"
    }
  }
}

SHA-256: def20bc5df10697616feda6d0649bbc3c26cb3496a4a840c1693c0fa5c4e1e08