DSH Trust Layer: four privacy-minimal observability plugins #1365
030611
started this conversation in
Show and tell
Replies: 1 comment
|
补充两张更具体的 Verification Receipt 使用图: 第一张由已发布的 dsh-verification-receipt@0.1.0 代码处理合成、非用户 DSH 事件后生成。左侧是仅在内存中参与分类的合成输入,右侧是实际 receipt 的部分落盘字段。隐私排除检查确认命令中的合成 token、workspace、工具输出和 call id 未进入 receipt。 再次强调:这些 receipt 只提供 execution-trace-only 的发现线索。它们不证明测试确实运行或通过,也不证明代码正确。欢迎反馈 verification heuristic 的误报、漏报,以及不同 DSH profile / rc 版本上的安装结果。 Repository: https://github.com/030611/dsh-verification-receipt |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment


Uh oh!
There was an error while loading. Please reload this page.
我做了四个可以独立安装的社区插件,尝试回答一个问题:怎样让 DeepSeek Harness 的运行过程更容易检查,同时避免把提示词、工具参数和结果正文复制进另一个日志系统?
四个独立插件
dsh-telemetry-redactor:在已配置遥测后端收到导出副本前,对受支持的凭证模式脱敏;不改写规范会话日志,也不承诺识别任意未知秘密格式。dsh-verification-receipt:每轮写入工具计数和词法层面的形似验证信号;不保存 prompt、工具参数或结果正文,也不证明测试真的运行或代码正确。qiushi-dsh-evidence-audit:写入本地哈希链 evidence receipts,用于发现部分编辑、删除和重排;它不是数字签名、可信时间戳或密码学认证。dsh-context-provenance:把公开运行时接口能够支持的结论明确标成Observed、Estimated或Unavailable;不返回提示词正文、消息内容、工具 schema 或原始路径。四个包都已发布 npm
v0.1.0,各仓库提供边界说明、测试与可复现的发布检查。它们可以单独使用,不要求一次安装整套。我特别希望得到三类反馈:
English summary: Four independently installable, community-maintained DSH plugins for privacy-minimal runtime observability: outbound telemetry-copy redaction, heuristic per-turn verification summaries, local hash-chained evidence receipts, and strictly labeled context provenance. They do not prove semantic correctness, provide signatures, or expose hidden prompts. Real compatibility reports and boundary-focused feedback are welcome.
Development and documentation were assisted by OpenAI Codex. The repositories disclose concrete test coverage and limitations; claims above were checked against their public manifests, source, and released
v0.1.0packages.All reactions