Original Summary

I found myself using the handoff skill too much — between sessions in different roles, and between different platforms (Codex and Claude for me). So, like programmers do, I wrote a solution for it: a shared memory for a fleet of agents, where each agent group has its own memory scope.<p>After some time using it, I saw the memory become a pile of junk, so I added a judge to each scope — an independent agent that acts as a gate to that scope&#x27;s memory. A contribution is admitted or declined, and the reason is recorded either way. Declines are kept, so a local console shows what an agent currently believes, where each belief came from, and what was kept out and why.<p>It ended up as a cross-platform governed memory system that works for me, so I open-sourced it.<p>It used to admit things it shouldn&#x27;t — in an early run (v1.12) it stored an irrelevant note about an office coffee machine. That&#x27;s fixed once a scope states its purpose (or has enough memory to imply one).<p>What it still doesn&#x27;t do, all stated in the README:<p>- A brand-new empty scope with no stated purpose still accepts that kind of note (<a href="https:&#x2F;&#x2F;github.com&#x2F;oren198&#x2F;Strata&#x2F;issues&#x2F;210" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;oren198&#x2F;Strata&#x2F;issues&#x2F;210</a>). - If you write a rule like &quot;support never stores customer account details&quot;, the judge doesn&#x27;t yet reliably use that rule to reject such notes when someone contributes them (<a href="https:&#x2F;&#x2F;github.com&#x2F;oren198&#x2F;Strata&#x2F;issues&#x2F;212" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;oren198&#x2F;Strata&#x2F;issues&#x2F;212</a>). On my adversarial test set — items written to trick the judge — the current build let 1 of 84 through; the early build let 2 of 72 through.<p>The judge model is set in config. I use qwen3-235b via OpenRouter, which I picked after trying six models on the same test suites; the numbers, and which cells are missing, are here: <a href="https:&#x2F;&#x2F;github.com&#x2F;oren198&#x2F;Strata&#x2F;blob&#x2F;main&#x2F;docs&#x2F;evidence&#x2F;judge-baseline-2026-09-20.md" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;oren198&#x2F;Strata&#x2F;blob&#x2F;main&#x2F;docs&#x2F;evidence&#x2F;ju...</a><p>Feel free to use it or contribute. I&#x27;d most like to hear where write-back breaks on someone else&#x27;s setup

  • 情报分类：技术学习与提效
  • 分类依据：内容涉及技术、AI、软件工具或工程实践
  • 信息来源：Hacker News 新项目
  • 发布时间：2026/9/23 04:27:18