Armature公司联合创始人Louis在Hacker News上介绍了一个为AI代理量身定制的新平台——Agent.reviews1。该平台采用类似G2和Trustpilot的模式,使AI代理能够在选择工具前阅读评测,使用后提交反馈1。通过npm CLI和API端点的集成,该平台已在50000多个AI代理会话中进行了测量1,目前已积累数千条评测记录1。
平台对人类用户和AI代理都免费开放,用户既可以查阅评测也可以发布评测1。为了确保数据质量,Agent.reviews采用了三层隐私保护机制,包括确定性规则过滤、Jev分类器和小型语言模型验证1。该平台已支持Claude Code、Codex和Cursor等多个AI代理平台1。
在实际运用中,平台已收集到有价值的反馈。Claude Code代理发现了Stripe SDK在缺少API密钥时会在健康检查页面上出现系统性崩溃1;同时有两个代理报告称Prisma即使在未连接数据库的情况下仍需要DATABASE_URL变量1。
Armature has introduced Agent.reviews, a review platform designed specifically for AI agents to evaluate and provide feedback on software tools.1 The platform functions similarly to existing services like G2 and Trustpilot, but tailored for autonomous agent workflows.1 According to Co-Founder Louis's announcement on Hacker News, Agent.reviews connects agents with software tools through an npm CLI and API endpoints, enabling agents to read reviews before selecting tools and submit feedback after use.1
The platform has already measured over 50,000 agent sessions and collected thousands of reviews.1 To maintain data integrity, Agent.reviews employs a three-layer privacy protection system consisting of deterministic rule filtering, a Jev classifier, and validation through a small language model.1 Both accessing and publishing reviews are available free of charge for both humans and agents.1 The platform currently supports agent platforms including Claude Code, Codex, and Cursor.1
Early use of the platform has surfaced specific technical issues. A Claude Code agent discovered that the Stripe SDK systematically crashes on its health check page when an API key is missing.1 Additionally, two separate agents reported that Prisma requires a DATABASE_URL variable even when no database connection has been established.1
评论
还没有评论,欢迎留下第一条。