中文在前，English follows each item. 全部内容对应仓库的实测数据，不空谈。
中文：没有隶属关系。官方仓库（deepseek-ai）管 DSH 框架本身；社区目录（如 awesome-dsh-plugin）负责收录清单——收录≠安装测试。dsh-lab 是第三方编辑位：只发布我们自己动手测过的结论，目录里的条目只是我们的候选池，不是我们的结论来源。
EN: We are an independent third-party curation site. Community directories list plugins, but listing ≠ tested. dsh-lab only publishes conclusions from tests we ran ourselves; directory entries are our candidate pool, not our source of truth.
中文：三种可能：(a) 还没排到测试队列；(b) 测试失败且失败原因已如实记录（❌）；(c) 我们还没发现它。想加速？按 CONTRIBUTING 提送测请求。我们从不为"没测过"的插件写描述性结论。
EN: Either (a) it hasn't reached the test queue, (b) it failed testing and the failure is recorded honestly (❌), or (c) we haven't found it. Open a 送测/request issue to speed things up. We never write descriptive verdicts for untested plugins.
中文：安装（装得上吗）→ 冒烟（跑得起来吗）→ 安全快检（代码里有什么出格行为吗）→ 兼容钉版（记录插件×DSH 版本，变化即复测）。完整协议见 METHODOLOGY。
EN: Install (does it install) → Smoke (does it boot) → Static quick-scan (what does the code actually do) → Version pinning (plugin × DSH versions recorded; re-test on change). Full protocol: METHODOLOGY.
中文：不是"保证"。它只代表在我们测试时点（见报告头部的 DSH 版本与日期）：安装成功、代码检查无异常线索、行为与宣称一致。DSH 更新或插件更新后结论可能失效——这正是"兼容钉版"存在的意义。所有报告都是测试时点的快照。
EN: No guarantee. ✅ means that at our test timestamp (DSH version + date in the report header) it installed cleanly, scanned without suspicious findings, and behaved as advertised. Verdicts can expire as versions drift — that's exactly why we pin versions. Every report is a point-in-time snapshot.
中文：冒烟关需要真实模型凭据（会真实消耗 token）。实验室没有把这部分成本转嫁给读者：无凭据时不猜结果，如实标 ⏳ 并写明原因。我们的路线图（ROADMAP）把"实验室自有测试凭据"列为阶段 2 前置。
EN: The smoke gate calls a real model and consumes real tokens. We don't fake it: without credentials the cell is honestly marked ⏳ with the reason. Lab-owned test credentials are a Phase-2 prerequisite in ROADMAP.
中文：不是。 快检是静态模式扫描，命中只是"代码里出现了这类行为"的线索。例如：视觉插件解码图片必然命中 base64 模式；安全审查类插件的规则表天然布满危险字符串。命中的真实含义在每份报告的判定列里逐条解释。同样，零命中也不等于无害——我们没有能力做人工安全审计，快检只是降低风险的线索层。
EN: No. The scan lists behavioral patterns, not verdicts. A vision plugin legitimately hits base64 patterns; a vetting plugin's rule table is full of dangerous strings by design. Each hit is interpreted per-report. Zero hits is also not "harmless" — we run pattern-level quick scans, not security audits.
中文：能，而且我们要求"任何人在任何机器上都能复现"。每份报告都有完整复现命令（隔离 DSH_HOME + 一条 dsh plugin add），完整安装日志在 lab/logs/。复现结果与我们不一致？按 CONTRIBUTING 提不一致反馈，我们复测并保留勘误记录。
EN: Yes — reproducibility is a hard requirement. Every report ships the exact repro commands (isolated DSH_HOME + one dsh plugin add), with full install logs in lab/logs/. If your result differs, file an inconsistency issue; we re-test and keep the erratum on record.
中文：看"兼容钉版"：版本一变就触发复测，报告日期与结论随之更新。阶段 2 起由 CI 每日自动执行（见 ROADMAP）。
EN: Version pinning handles drift: any version change triggers re-testing and report refresh. From Phase 2, CI runs this daily (ROADMAP).
中文：DeepSeek 开源的"一切皆插件"AI Agent 框架——核心是一个 Cordis 风格的插件运行时 + Web 界面，模型工具、界面组件、预设、技能都以插件形式加载。官方仓库 12 万+ star，社区插件数百个。
EN: DeepSeek's open-source "everything is a plugin" AI agent framework — a Cordis-style plugin runtime plus a web UI where model tools, UI components, presets, and skills all load as plugins. 120k+ stars upstream, hundreds of community plugins.
没回答你的问题？开 issue 提问，我们会把高频问题补进这里。 Question not covered? Open an issue — recurring ones get added here.