TapeId 独立复现两份 TAP 草案:13 处歧义TapeId re-implements two TAP drafts, flags 13 ambiguities
TapeId_AI 表示,团队仅依据 TapeOut 两份 TAP 协议草案的规范文字,各自重新实现了一遍,再用草案自带的测试向量验证:#84 花钱规则电路(Policy Circuit Gate)通过 235/236 项检查,#83 代理护照(Agent Passport)通过 182/182 项。过程中发现 13 处规范存在两种理解方式,已写成报告贴到两个 PR 下。团队称,这样做的目的是在草案阶段暴露文字歧义,避免钱包与 AI 代理接入后再修改。TapeId_AI said it re-implemented two TapeOut TAP drafts from the spec text alone and ran them against the drafts' own test vectors: #84 Policy Circuit Gate passed 235 of 236 checks, and #83 Agent Passport passed 182 of 182. The team found 13 places where the text can be read two ways and posted reports under both PRs, aiming to expose spec ambiguities while still in draft rather than after wallets and AI agents have adopted them.
来源 · 2 条Sources · 2
- @TapeId_AI
我们把 TapeOut 的两份协议草案(TAP)只照着规范文字重新实现了一遍,再拿草案自带的测试数据来跑: #84 花钱规则电路(Policy Circuit Gate):235/236 通过 #83 代理护照(Agent Passport):182/182 通过 过程中找到 13 处规范可以有两种理解的地方,已经写成报告贴在两个 PR 下面。 为什么要再做一遍? 一份规范有没有用,要看两个互不认识的人各自照着它写,能不能得到一样的结果。两个验证器读同一段文字却得出不同结论,就说明文字有漏洞。趁还是草案的时候找出来,比等钱包和 AI 代理都用上了再改便宜得多。 所以我们没看参考代码,
- @TapeId_AI
We implemented two TapeOut TAP drafts a second time, from the spec text alone, and ran them against the drafts' own test vectors. #84 Policy Circuit Gate: 235/236 checks pass #83 Agent Passport: 182/182 checks pass Along the way we found 13 places where the text can be read two ways. We posted the