把一个"两派都有道理、再论证下去没有新信息"的架构/设计之争，转成一个能出判决的可测量实验。写实验规格（事前判据 + 判决树 + kill criterion + 时间盒）、审已有规格的漏洞、把结论回填。当出现「A 方案 vs B 方案该选哪个」「这条路会不会无界增长」「这个假设成不成立」而争论已达推理极限时使用；也用于 review 别人写的实验规格 / 研究轨文档。不用于普通 benchmark、bug 复现、性能调优。
把一个"两派都有道理、再论证下去没有新信息"的架构/设计之争，转成一个能出判决的可测量实验。写实验规格（事前判据 + 判决树 + kill criterion + 时间盒）、审已有规格的漏洞、把结论回填。当出现「A 方案 vs B 方案该选哪个」「这条路会不会无界增长」「这个假设成不成立」而争论已达推理极限时使用；也用于 review 别人写的实验规格 / 研究轨文档。不用于普通 benchmark、bug 复现、性能调优。
Make a Windows binary start on older Windows (Server 2016 / Windows 10 1607) and stop depending on the VC++ redistributable. Diagnose "entry point not found" and missing-DLL loader failures, decide which APIs must be resolved with GetProcAddress, and know…
Install CLI tools on Windows and make them immediately available in new terminals without logout. Avoid the PATH propagation traps that waste time.
Build, install, launch, and diagnose AgenTerm from a local macOS checkout. Use when a local AgenTerm build must appear as a native Dock application, when Finder opens a bare agenterm binary through Terminal with Last login output, or when validating the local…
Launch, attach, diagnose, and clean AgenTerm Windows GUI/server instances without killing work or lying about visibility. Use when the user cannot see AgenTerm windows, GUI dies after agent launch, multi-instance main/dev/work attach is needed, dual/stale…
Operate and diagnose AgenTerm's exact-SHA Release Candidate and approved byte-promotion workflows. Use for Candidate creation, release rehearsal, GitHub Actions monitoring, release authentication or dispatch failures, Promotion approval, tag/Release…
Work safely in the AgenTerm repository while other agents commit to the same checkout, and judge test-failure and "pre-existing flake" claims correctly. Use when dispatching subagents across platform swimlanes, when committing alongside concurrent agents, or…