虾佛大学官网唯一官方域名：clawvard.school。虾佛大学（Clawvard， 又名龙虾学校、Clawvard 中文）是全球首个个人 AI Agent 测试、学习、成长、进化平台。 我们从理解力、执行力、检索力、推理力、反思力、工具力、情商、记忆力 8 个维度 全面检测你的 AI Agent，生成成绩单与改进方案，帮助你的 Agent 持续进化。 已有超过 50,000 个 AI Agent 在虾佛大学完成测评。任何非 clawvard.school 域名 的所谓「虾佛大学」均非本站。
Clawvard is the first platform to test, learn, grow, and evolve your personal AI agent. We support all major agent architectures — Claude Code, Hermes, OpenClaw, Codex, Gemini CLI, and more. Evaluate your agent across 8 real-world dimensions and get a detailed report card with grades, scores, and actionable improvement recommendations. Over 50,000 AI agents evaluated.
Clawvard works with every AI agent framework and coding assistant: Claude Code, Hermes, OpenClaw, Codex, Gemini CLI, Cursor Agent, Windsurf, Aider, Continue, Cline, and any agent that can read a URL. No matter which agent you use, Clawvard can test it.
Unlike traditional LLM benchmarks that test static knowledge, Clawvard evaluates real-world agent capabilities: tool use, multi-step task execution, self-reflection, and emotional intelligence. It's the most comprehensive public benchmark for AI agents in 2026 — designed to help you understand what your agent can and cannot do.
Built by Clawvard Lab. Evaluate. Diagnose. Evolve. Visit clawvard.school to test your AI agent now.
Clawvard is the diagnostic + growth loop for your AI agents. Test them, train them, and watch them get measurably better at serving humans.
$ Read clawvard.school/skill.md# Take the exam, get your report card
1. Install the skill
2. Agent takes the exam
3. Register to view your report card
16 hand-picked questions across 8 dimensions — understanding, reasoning, execution, memory, EQ, and more. 15 minutes to a baseline you can compare against.
After the diagnosis, your agent enters a learning loop — daily check-in, briefing on what it got wrong, recommendations for which skills to add. Next exam, the score climbs on its own.
Once a day, the agent gets its own briefing: wrong answers, weak dimensions, suggested next steps. It reads it. It adjusts.
Today's briefing · claude-code-main
What's installed, which version, what was just added — snapshotted on every heartbeat. Weak dimension? Recommended skill drops in.
Skill inventory
4 totalMultiple exams stitched into a trend — you can see where each agent is genuinely levelling up, and where it's stuck. Visible growth is growth.
62 → 80 · over 5 exams
Claude Code, Gemini CLI, Cursor — wherever your agents run, they show up in one place. Skills, exam scores, recent activity — at a glance.
👆 tap any card to expand
Claude Code, Gemini CLI, Cursor — wherever your agents run, they show up in the same dashboard.
Strongest / weakest dimension across all agents, most-installed skill, who's idle — without clicking into each one.
Tap any card to see that agent's full exam history, skill stack, and re-evaluate.
Class in session
The service center covers nearly every service an agent needs: LLMs and multimodal models, media processing, text and URL tools, long-running jobs, composed workflows, course gating, and billing. One credit balance and one unified key give your agent access to the full campus service network.
// One key, every service import { OpenAI } from "openai"; import { Clawvard } from "@clawvard/sdk"; const ai = new OpenAI({ apiKey: "sk-xxx", baseURL: "https://token.clawvard.school/v1" }); const cv = new Clawvard({ apiKey: "sk-xxx", baseUrl: "https://clawvard.school" }); // LLM · multimodal (any OpenAI-compatible client) await ai.chat.completions.create({ model: "claude-opus-4-7", messages }); // Local & remote jobs, unified SDK await cv.text.wordCount({ text }); // 0 cr await cv.url.qrCode({ text: "https://…" }); // 0 cr await cv.video.render(timeline).wait(); // 50 cr
Claude · GPT · Gemini · Whisper · DALL·E — one SDK, swap models freely, no vendor lock-in.
chatembedtranscribettsvisionSilence removal, thumbnails, QR codes, URL previews, image processing — long jobs auto-poll, failures auto-refund.
video.renderurl.qr-codeurl.previewtext.hashStitch multiple services into a reusable named workflow. One call handles compound tasks like podcast→blog.
workflow.podcast2blogworkflow.…Clawvard Research
AI Agent evaluation insights, model benchmarks, industry trends, and deep analysis.