.codex-plugin/plugin.json

{
  "name": "pagent",
  "version": "1.0.0",
  "description": "pagent is the AI Experiment Engineer for live websites. It helps authenticated users inspect tests, results, hypotheses, audiences, and configuration, then manage permitted experiment workflows from ChatGPT while preserving pagent permissions and review controls.",
  "author": {
    "name": "pagent",
    "url": "https://pagent.ai"
  },
  "homepage": "https://pagent.ai",
  "mcpServers": "./.mcp.json",
  "interface": {
    "displayName": "pagent",
    "shortDescription": "Optimize website experiments",
    "longDescription": "pagent is the AI Experiment Engineer for live websites. It helps authenticated users inspect tests, results, hypotheses, audiences, and configuration, then manage permitted experiment workflows from ChatGPT while preserving pagent permissions and review controls.",
    "developerName": "pagent",
    "category": "Business & Operations",
    "capabilities": [
      "Read website experiment results",
      "Manage hypotheses and tests",
      "Configure website targeting and tracking"
    ],
    "websiteURL": "https://pagent.ai",
    "supportURL": "https://www.pagent.ai/contact/",
    "privacyPolicyURL": "https://www.pagent.ai/privacy-policy/",
    "termsOfServiceURL": "https://www.pagent.ai/mcp-terms",
    "defaultPrompt": [
      "Which of my running website tests is winning, and how strong is the evidence?",
      "Is any active test unreliable or short on traffic?",
      "Show the hypotheses I could turn into my next experiment."
    ],
    "composerIcon": "./assets/icon.png",
    "logo": "./assets/icon.png"
  },
  "extensions": {
    "com.openai": {
      "review": {
        "test_cases": {
          "positive": [
            {
              "description": "List running website tests and identify their pages.",
              "prompt": "List my running tests and tell me which page each one is on.",
              "expected_behavior": "Returns the running tests with their identifiers, status, page, and currently allowed next actions without inventing results.",
              "tools_triggered": "list_websites, get_tests"
            },
            {
              "description": "Explain a conclusive test result with its measured evidence.",
              "prompt": "How did the annual billing toggle test finish? Include conversion rates, uplift, and the verdict.",
              "expected_behavior": "Returns the available result with control and variation conversion rates, relative uplift, probability to beat control, and the conclusive verdict.",
              "tools_triggered": "list_websites, search_entities, get_test_results"
            },
            {
              "description": "Surface a sample-ratio mismatch instead of trusting an apparent winner.",
              "prompt": "Is the homepage savings-headline test reliable enough to trust?",
              "expected_behavior": "Explains the observed result while prominently warning that the sample-ratio mismatch makes the evidence unreliable.",
              "tools_triggered": "list_websites, search_entities, get_test_results"
            },
            {
              "description": "Capture a new hypothesis for an existing website page.",
              "prompt": "Save this hypothesis for the pricing page: replacing technical plan names with outcome-based names will improve plan selection.",
              "expected_behavior": "Creates one editable pricing-page hypothesis and returns its identifier, target page, and whether an equivalent hypothesis already existed.",
              "tools_triggered": "list_websites, search_entities, capture_hypothesis"
            },
            {
              "description": "Pause one exact live test after checking lifecycle constraints.",
              "prompt": "Pause the running trust-badges test on the product page.",
              "expected_behavior": "Identifies the exact test, obtains user approval for the live-site change, pauses only that test, and reports its new status.",
              "tools_triggered": "list_websites, explain_lifecycle, search_entities, pause_test"
            }
          ],
          "negative": [
            {
              "description": "Do not invoke pagent for standalone copywriting.",
              "prompt": "Write three alternative headlines for a generic productivity app homepage.",
              "expected_behavior": "The app should not be invoked because the request does not ask to read or manage a pagent workspace."
            },
            {
              "description": "Do not invoke pagent for unsupported third-party analytics.",
              "prompt": "Open my Google Analytics account and show last month's acquisition channels.",
              "expected_behavior": "The app should not be invoked because pagent does not provide access to Google Analytics accounts."
            },
            {
              "description": "Do not approximate unsupported account-wide deletion.",
              "prompt": "Delete all organisations, websites, tests, and customer data in my pagent account.",
              "expected_behavior": "The app should not be invoked because no published tool supports account-wide customer-data deletion."
            }
          ]
        },
        "demo_recording_url": "https://drive.google.com/file/d/1C6LtsXtJsSadmqMT8oRpK639ehRX4I34/view?usp=sharing"
      },
      "publication": {
        "release_notes": "Initial ZIP submission of the pagent MCP integration, including website experiment workflows, listing information, reviewer scenarios, and the existing walkthrough video."
      }
    }
  },
  "apps": "./.app.json"
}