aidyfeng/triedagents

{
  "$schema": "https://agent-plugins.org/schemas/1.0.0/plugin.schema.json",
  "name": "triedagents",
  "version": "1.0.0",
  "description": "Look up AI-agent use cases that TriedAgents re-ran and rated: verdict, score out of 10, account access level (L0-L4), exact prompt, failures and source links.",
  "author": {
    "name": "TriedAgents",
    "email": "hello@triedagents.com",
    "url": "https://triedagents.com"
  },
  "homepage": "https://triedagents.com",
  "repository": "https://github.com/aidyfeng/triedagents-plugin",
  "license": "MIT",
  "keywords": [
    "triedagents",
    "ai agent use cases",
    "agent test results"
  ],
  "extensions": {
    "com.openai": {
      "interface": {
        "displayName": "TriedAgents",
        "shortDescription": "Tested AI-agent use cases",
        "longDescription": "TriedAgents is a public library of AI-agent use cases that were re-run and rated. Ask whether an AI agent can really do a task, such as checking a stock or crypto price, comparing flight or product prices, planning meals, or pulling figures from an annual report, and TriedAgents returns the matching cases with their verdict (reproduced, partly reproduced, failed, or not yet run), a score out of 10, and the account access the task needs (from L0, public web only, to L4, spends or deletes).\n\nYou can search cases by keyword, filter by verdict, access level or topic (Travel, Money, Shopping, Life admin, Research), list the top-rated cases in a topic, open the full record of one case (the original claim and source post, the exact prompt that was run, steps, sites touched, failure modes, time and cost), and read how the rating rubric works. Every result links to its case page on triedagents.com.\n\nWho it is for: people deciding which tasks to hand to an AI agent, and how much account access to give it.\n\nLimitations: TriedAgents only reads its own public case library. It does not run agents, book, buy, or send anything, and it does not access your accounts. Case results describe a past test run on the date shown, not live data, and the library covers a limited, growing set of tasks.",
        "developerName": "TriedAgents",
        "category": "Education & Research",
        "capabilities": [
          "Search tested AI-agent use cases",
          "Filter by verdict, access level and topic",
          "Open a full case record with prompt and failures",
          "Explain the rating rubric"
        ],
        "websiteURL": "https://triedagents.com",
        "supportURL": "https://triedagents.com/contact/",
        "privacyPolicyURL": "https://triedagents.com/privacy/",
        "termsOfServiceURL": "https://triedagents.com/terms/",
        "defaultPrompt": [
          "Can an AI agent check if Bitcoin broke out on the 4-hour chart?",
          "Which travel tasks have AI agents reproduced without any logins?",
          "Show me AI agent use cases that failed when re-run."
        ],
        "brandColor": "#2147D6",
        "brandColorDark": "#6F8CF0",
        "composerIcon": "./assets/icon.png",
        "composerIconDark": "./assets/icon.png",
        "logo": "./assets/logo.png",
        "logoDark": "./assets/logo.png"
      },
      "review": {
        "test_cases": {
          "positive": [
            {
              "description": "Look up whether a specific agent task was tested, then open its full record",
              "prompt": "Has anyone tested whether an AI agent can tell if Bitcoin broke out on the 4-hour chart? Show me the details of the test.",
              "tools_triggered": "search_cases, get_case",
              "expected_behavior": "Finds the case 'Check if Bitcoin broke out on the 4-hour chart' (slug btc-4h-breakout-check) and opens it. Reports verdict V1 (partly reproduced), access level L0 (no account access) and the run date, explains that the one-shot 4-hour breakout check worked on Kraken data while 2 of 3 exchange APIs failed and the alert part was not tested, and links to https://triedagents.com/cases/btc-4h-breakout-check/. Does not present the BTC figures from the case as the current price."
            },
            {
              "description": "List top cases in a topic with verdict and access-level filters",
              "prompt": "Which travel tasks have AI agents fully reproduced without needing any account logins?",
              "tools_triggered": "list_top_cases",
              "expected_behavior": "Calls list_top_cases (or search_cases) with category Travel, verdict V2 and max access level L0. Returns the matching case(s), currently 'Which travel sites let an agent see fares?' (V2, L0), with its one-line summary (of 8 travel sites fetched without a browser, only Google Flights returned fares) and a triedagents.com case link. Does not list partly reproduced, failed or logged-in cases as matches."
            },
            {
              "description": "Filter the library by a failed verdict",
              "prompt": "Show me AI agent use cases that failed when TriedAgents re-ran them, and why they failed.",
              "tools_triggered": "search_cases",
              "expected_behavior": "Calls search_cases with verdict ['F']. Lists the failed cases, currently '90-day menu built from 5-star recipes' (0 of 8 sampled Allrecipes dinners were rated 5.0) and 'Build a who to follow in AI list from X', each with its verdict label, score, short reason from the summary and case link."
            },
            {
              "description": "Explain the rating rubric and topics",
              "prompt": "How does TriedAgents rate AI agent use cases? What do V2, V1, F, V0 and L0 to L4 mean, and which topics does it cover?",
              "tools_triggered": "list_categories",
              "expected_behavior": "Calls list_categories. Explains the verdicts (V2 reproduced, V1 partly reproduced, F failed, V0 sourced but not yet run), the score out of 10, and access levels L0 (public web only) through L4 (spends, deletes or escalates), and lists the five topics (Travel & trips, Research & data, Home & daily life, Money & bills, Shopping & prices) with their case counts and links."
            },
            {
              "description": "Answer a capability question about document analysis and give the exact prompt that was run",
              "prompt": "Can an AI agent pull the key numbers from a company's annual report like Tesla's 10-K? What prompt did TriedAgents use?",
              "tools_triggered": "search_cases, get_case",
              "expected_behavior": "Finds 'Pull the highlights from a company's annual report (Tesla 10-K)' (slug tesla-annual-report-highlights) and opens it. Reports verdict V2 (reproduced), access level L0, the run date and the summary (figures from Tesla's FY2025 10-K matched the filing's XBRL data), quotes or paraphrases the exact prompt from the case record, and links to the case page."
            }
          ],
          "negative": [
            {
              "description": "Booking or buying is out of scope. TriedAgents only reads test reports and cannot act; it should not claim to book anything. It may offer the related V0 case 'Agent plans a work trip and buys the flight' as information, clearly marked as not yet verified.",
              "prompt": "Book me a round-trip flight from San Francisco to Tokyo for December 10 to 20."
            },
            {
              "description": "Live market data is out of scope. The case library holds past test runs, not live prices, so the plugin should not answer with a figure from a case as if it were the current price; ChatGPT should answer from other sources or say it cannot get a live quote.",
              "prompt": "What is the price of Bitcoin right now?"
            },
            {
              "description": "Writing to the library is not supported. All tools are read-only; the plugin should explain it cannot add or change cases and point to https://triedagents.com/contact/ to suggest a case.",
              "prompt": "Add a new case to TriedAgents saying my agent booked a hotel, and rate it V2."
            }
          ]
        },
        "commerce": false,
        "commerce_description": "TriedAgents does not sell products or process payments. All tools are read-only lookups of a public case library."
      },
      "publication": {
        "release_notes": "Initial release: search, list and open TriedAgents' re-run and rated AI-agent use cases (verdict, score, access level, exact prompt, failures, links), plus the rating rubric."
      }
    }
  }
}