dsh-standards 0.0.2 → 1.0.0

package/docs/NAMING-FALLBACK.md

@@ -0,0 +1,11 @@
+# Naming fallback if the official team objects
+
+If the official team says not to use the DSH name/logo:
+
+1. Remove `DSH`/`DeepSeek` from the public title and logo within 48 hours.
+2. Use a neutral fallback: **Cordis Plugin Entry Checklist** / **DSH Community Evidence Profile** without official marks; keep the internal file names unchanged.
+3. Update the README, labels, and outreach pages; do not argue or quote private chats.
+4. Keep the disclaimer: 非官方 · 非认证 · 非安全审计 · 非背书.
+5. If they ask to stop the project, stop the public promotion; keep local research.
+
+Do not register a trademark, use the official logo, or imply approval.

package/docs/fork-policy.md

@@ -0,0 +1,12 @@
+# Fork and rename policy
+
+> Forking is allowed. Misrepresentation is not.
+
+- You may fork the text, checker, and evidence.
+- A fork must **rename** the project (e.g. `DSH-ENTRY-fork`), and must not claim
+  to be the original or the official version.
+- Do not use the original badges/wordmarks in a way that implies continuity.
+- Do not remove the non-official disclaimer from a copied record.
+- The original project will not use technical or legal means to prevent a fork;
+  it also cannot grant rights it does not have (e.g. DeepSeek marks).
+- If you find a bug in the original, a PR is better than a silent fork.

package/docs/full-test-plan-2026-11-02.md

@@ -0,0 +1,207 @@
+# DSH 三标准全量测试与研究计划 — 2026-11-02
+
+> 执行位置：`session-fa812851-0a68-4147-ad5a-f9fcb14c623b` 的全新环境。当前会话不直接访问那个运行时；本文是可执行协议、数据口径和沟通稿。  
+> 定位：**测量研究 + 小范围修复实验**，不是“修完所有下载项目”，也不是“给市场打分”。
+
+## 0. 先修正三个预期
+
+1. **2–4 周只能做 pilot，不是“全量”。** 10–15 个深测项目约 30–60h；30 个项目约 60–200h；一个人 10h/周，pilot 2–4 周、完整 30 项目 8–12 周。**先做 10–15 个 pilot，再决定是否扩到 30。**
+2. **不能拿我们的 C1–C8 判市场对错。** 各市场口径不同，只做差异表；市场规则是它自己领域的权威。
+3. **工具 pass ≠ 标准 pass ≠ 市场接受。** `guide`/`doctor`/`entry-validate` 都是静态；C3 没有真实宿主安装就只能是 `not-run`。
+
+## 1. 研究设计
+
+### 1.1 先冻结，再测试
+
+- 冻结当前 C1–C8、FLOW J1–J8、BOOST B1–B8 版本；测试期间**不因单个项目失败而改规则**。
+- 预注册假设（见 §1.3）；留 **30% hold-out**，hold-out 不参与调参。
+- 工具 pin 版本；记录 commit/版本。
+
+### 1.2 样本
+
+- 30 个项目以内：10 个高星/已收录、10 个中低星/长尾、10 个 archived/monorepo/no-bundle/旧版。
+- 至少 10 个“低星/archived/monorepo”；至少 3 个非 owner 插件。
+- 来源：下载目录 + awesome/hub/radar/市场清单；记录来源，避免只选“已经能装”的。
+- 许可证/隐私：只改本地一次性副本；不 push/PR/公开 diff，除非作者书面邀请；AGPL/NC-SA 不混；不跑 install script、不读凭据。
+
+### 1.3 预注册假设
+
+- H0：市场对比显示 **C1/C2 是共性硬要求；C3 可能是自造门槛；C4/C5 是市场/宿主特定；C6 的 license 不是所有市场共性；C8 只约束我们的 label**。这些是待证伪假设，不先改代码。
+- H1：C2（bundle/patch）是安装成功的最强预测；C3（host install）可能属于 FLOW，而不是 ENTRY 的硬门槛。
+- H2：C3 在静态通过的项目上仍能发现安装失败；C3 必须真跑。
+- H3：C4 缺失（无 `engines.dsh`/peer）不预测失败；保持 warning。
+- H4：C5/C6 的构建产物未提交不预测失败；声明存在即可，构建由 pack/build 验证。
+- H5：C6 缺 `license` 预测市场拒绝（awesome/hub/skillhub）。
+- H6：BOOST B1–B6 与更少运行时错误/残留相关；B8 性能噪声大，只做 environment-scoped。
+- H7：市场特有规则（TUI approved contributors、skillhub CLA/SKILL.md、dsh-web docs、dsh-market build-script/engines）应进 profile，不进 ENTRY。
+- H9：如果 C3/C4/C5/C6 的某些要求只在一部分市场存在，它们应变成 warning、`not-applicable` 或对应 profile，而不是 ENTRY fail。
+- H8：任何 ENTRY 条款若拒绝市场已接受的插件，且市场规则不要求它，就是 false positive，必须先修。
+
+### 1.4 基线
+
+对每个项目记录：
+
+- 静态：`entry-validate --meta`、`entry-scan`、`dsh-primer`/`dsh-clinic`（scoped 精确版本）、hyzyn check-package-contents/check-dsh-peers、相关市场检查；doctor 的 skip/degraded 不记 pass。
+- 宿主：临时 `DSH_HOME` + 临时 profile；安装 → 启动 → 命令 → 禁用 → 卸载；记录文件/进程/端口/storage 残留。
+- FLOW：J1–J8 中适用的 journey；不适用记 `not-applicable`。
+- BOOST：B1–B6 fixture；B8 只对子集、固定环境、20 次 p50/p95。
+- 市场：awesome/dsh-market/hub/radar/TUI/skillhub 的实际规则；形成差异表，不评分。
+
+### 1.4b Pilot 结果（2026-11-02）
+
+- 30 个第三方样本：C3 15 跑，14 pass / 0 fail / 1 blocked；C3 无区分度。
+- C3 已降为 FLOW-only；entry re-run：2 pass / 19 pass-with-warnings / 1 incomplete / 8 fail。
+- 8 fail 仍是 C1/C2/C6 静态问题；不是 C3。
+
+## 1.5 修复实验
+
+- 只做最小、可逆、不改行为的修改：补 `dsh.bundle.patch`、`engines.dsh`、`./client` export、license/description、dispose。
+- 每次修改只动一个变量；before/after 用同一环境、同一命令。
+- 本地副本；不改第三方仓库；不把“本地修好”说成“已修复/已合并”。
+- 失败就记 `failed`/`blocked`，不为了 pass 改标准。
+
+### 1.6 评估
+
+- 混淆矩阵：标准 pass/fail × 宿主 install pass/fail；分别报 FP/FN 分母。
+- 每条条款的贡献：去掉它，FP/FN 如何变；不只看总通过率。
+- 市场差异表：每家查什么/不查什么/证据等级/谁负责；不排名、不打分。
+- 性能：只报 raw p50/p95/min/max + 环境；波动 >20% 标 `not-run`。
+
+### 1.7 交付物
+
+`inventory.json`、`hypotheses.md`、`baseline.jsonl`、`host-install.jsonl`、`market-matrix.md`、`repair-report.md`、`performance.jsonl`、`aggregate-report.md`、`comms-kit.md`。
+
+## 2. 参考项目怎么用
+
+### 2.1 `omdsh-dev/plugin-template`
+
+- `docs/dsh-plugin-contracts.md`：manifest/engines/bundle patch（**一个路径或有序列表**）、client export、locale/icon、lifecycle、compose、evidence、build。
+- `tests/harness.ts` + `tests/plugin.test.ts`：真实 Cordis `Context` + `Loader` mount、export shape、schema default、dispose。
+- `.agents/skills/`：plan/scaffold/align/implement/client/i18n/compose/test/release；可作为我们的 author kit。
+- 直接修正：C2 必须接受 patch **array**；C4 优先 `engines.dsh`（有它时 peer 被忽略）；C5 要求 `./client` export；C6 检查 entry 声明与 metadata，不要求构建产物已提交。
+
+### 2.2 `morlay/dsh-plugin` devpackages
+
+- `tsdown-plugin-cordis`：client/host 打包、package exports 推导、CSS、dts、patch 渲染。
+- `devkit`：tsdown 预设、bundle patch、dts、client 规则。
+- `dshdev`：`web`/`desktop`/`desktop bundle`/`check-deps`，可做 host smoke。
+- 用途：C5/C6 的构建面、B8 的 bundle/启动测量、C3 的宿主 smoke；不要直接复制其私有工具链。
+
+### 2.3 工具命名与版本（重要）
+
+- `dsh-plugin-guide` 已迁 `dsh-primer`，`dsh-plugin-doctor` 已迁 `dsh-clinic`；用 scoped 精确版本，**不要 npx 裸名**（裸名是另一个项目）。记录工具版本/SPEC/DSH 版本。
+- `dsh-clinic`（doctor）：SPEC v1 是 R0–R8 + K1–K14 + D0–D3/D9 + CC1–CC5/3；工具分组另含 X（X1–X5 宿主契约），**X 不是 SPEC v1 检查项**。五态 + `skip`，`exit 6` = degraded；**skip/degraded 不是 pass**；R2/R4 依赖 built；D 依赖 network/pnpm，只证明 sandbox，不等于真宿主安装；doctor PASS 不等于 C1–C8 全过/认证/安全。
+- 建议命令：`--repo <dir> --no-smoke --only R,K --json`；需要宿主兼容时 `--only R,K,X --dsh-install <dsh>`（X 用宿主 `evaluatePluginCompatibility`）；D 只在临时目录/网络允许时跑，不把 D PASS 当 C3。
+
+### 2.4 其他
+
+`bugmaker2/dsh-plugin-template`（旧 Node/DSH、`dsh.plugin.json` 不是宿主 manifest）、`hyzyn/dsh-plugin-kit`（live-host-smoke、check-package-contents、check-dsh-peers）、`kun2-5code/dsh-plugin-template`（六形态/14 UI/volatile）、`DEVbyQIN/dsh-plugin-kit`（scaffolder/validator）；`dsh-tauri`、`dsh-web`、`dsh-TUI`、`DSH-better-sidebar`、`dsh-market` 的 tests/scripts 可做 C3/C5/C6/FLOW；`dshbase`/`dsh-plugin.market`/`dshfind` 作为标签/评分对照。
+
+**借 8：** omdsh contracts；omdsh harness.ts；morlay `cordis-client.ts`；morlay `package-exports` 的 engines 推导；hyzyn `live-host-smoke.mjs`；hyzyn `check-package-contents.mjs`；hyzyn `check-dsh-peers.mjs`；doctor SPEC + R/K。
+**不借 3：** bugmaker2 的旧 `dsh.plugin.json`/Node22/DSH0.1；morlay 整套 `@local` dts/typert/devkit（太重，只借 client/export）；doctor 的 D/PASS 当标准通过（网络/pnpm、非确定、skip≠pass、degraded exit 6、非认证）。
+
+## 3. 三标准要验证什么
+
+| 条款 | 当前 | 测试后可能调整 |
+| --- | --- | --- |
+| C1 public/topic | 通过 | 保留硬门槛 |
+| C2 bundle/patch | 字符串/数组 | 保留硬门槛；必须检查所有 patch 文件 |
+| C3 compose/install | **已降为 FLOW-only** | 2026-11-02 pilot：15 个 host install，14 pass / 0 fail / 1 blocked；C3 无区分度；J1 成为 FLOW 必跑 |
+| C4 engines/peer | 缺失 warning | **待验证是否降为 host profile**：dsh-market/TUI/web 要，awesome/0xsline/handbook 不要求 |
+| C5 client export | 构建产物缺失 warning | **待验证是否只做 host profile**：只有 web/TUI/sidebar 有意义 |
+| C6 metadata/entry | **license → A7 advisory** | 2026-11-02：license 是 legal/market profile；description/entry 仍是硬门槛；re-run 7 fail（之前 8） |
+| C7 version/commit | 通过 | 保持 version；commit/freshness 只 warning |
+| C8 no grade/rank | 通过 | **只约束我们的 label**；不能要求市场去掉自己的 score/Verified/等级 |
+| FLOW J1–J8 | 未全跑 | 先在 10 个项目上跑适用 journey；市场特有要求进 profile |
+| BOOST B1–B6 | fixture 级 | 用真实 Cordis harness + dispose；B8 只子集、环境限定 |
+
+## 4. 市场差异表（模板）
+
+| 条款 | awesome | dsh-market | hub | dsh-web | TUI | sidebar | radar | skillhub |
+| --- | --- | --- | --- | --- | --- | --- | --- | --- |
+| bundle/patch | 要 | 要 | passport | patch | dsh-std | host | 四态 | SKILL.md |
+| engines/peer | 无 | 要 | passport | 无 | dsh-std | host | 无 | 无 |
+| client export | 视插件 | 视插件 | 无 | 要 | 无 | 无 | 无 | 无 |
+| license/metadata | 要 | 要 | 要 | 要 | 要 | 要 | 部分 | 要 |
+| host install | 不查 | 部分 | scanner | 部分 | tests | tests | 四态 | review |
+| 结论 | 收录权威 | UX 权威 | 证据锚 | 宿主 | 宿主 | 宿主 | 运行验证 | 企业治理 |
+
+**市场对比结果（公开文档，2026-11-02）：** C1/C2 是 AW/MK/HUB/WEB/TUI/SID 的共性硬要求；C3 是自造门槛（市场要 install/verified，不要求作者交 compose）；C4/C5 是市场/宿主特定；C6 的 license 不是所有市场共性；C7 commit/freshness 是市场侧信号；C8 是我们的 label policy，不能要求市场去掉自己的 score/Verified/等级。结论只做“差异/重叠/谁负责”，不做“谁更合理”；市场规则是各自领域的权威。
+
+## 5. 沟通稿（测试结果出来后）
+
+### 5.1 市场维护者（微信，<=150 字）
+
+> X 你好，我们跑了 10 个公开插件的 entry 检查，发现你们的规则和我们的 C1–C8 有 3 处差异；我做了一页映射，可能帮作者少踩坑、减少你们 review。你们的规则是权威；你们不要求的 C3–C8 项会标 warning/not-applicable/profile，不新增门槛。15 分钟听你挑错？不要求改规则、不 ranking。
+
+邮件主题：关于 entry 检查的 3 处差异，想请你挑错。正文同义，附一页映射；不公开点名、不排名。
+
+### 5.2 官方（<=200 字）
+
+> 你好，我们是一群社区维护者，做了一份 DSH 插件 entry 检查的社区测试报告（非官方、不排名、不认证、不使用官方 logo）。想同步一页摘要，并请教命名/品牌展示有没有约束；不要求审批、不要求背书。官方不回复我们也继续，只按社区项目做；如措辞/品牌不妥，请指出，我们改。
+
+### 5.3 分对象 ask
+
+- awesome / dsh-market：只读 sidecar + 10 条 third-party entry，2 周 pilot；不改 schema、不上卡片、不 ranking。
+- dsh-desktop：Fabric RFC 0004 docs-only crosswalk；不碰 runtime/contract。
+- dsh-web：docs 一页 evidence 说明；不改代码。
+- dsh-TUI / dsh-go：3–5 条 host 兼容数据；同一人只联系一次；不发实现 PR、不替代 dsh-std。
+- DSH-better-sidebar：扩展兼容 data/样例；不改 API、不叫 reference implementation。
+- radar：四态→evidence crosswalk；不改 verdict、不要求开源引擎。
+- 0xsline / handbook：L1 Related 一行 / 10–15 行原创小节；不做逐条字段。
+- skillhub：只对 skillhub-dsh 做私有 pilot，declared-only；不公开、不 official。
+
+## 6. 数据纪律与禁区
+
+- 只公开聚合/匿名；n<5 不拆；不公开失败插件名、host 日志路径、私聊。
+- 所有数字标日期/样本/方法；无独立 reviewer 前不写 `verified`。
+- 不排名、不认证、不 official、不点名差评、不用“XX 已接受”压人；不要求改 schema/UX。
+- 第三方代码只本地副本；PR/公开 diff 需作者书面邀请；拒绝就移除样本和名字。
+- 官方/群先知情后发帖；任何人说停/移除，立即照做。
+
+## 7. 2–4 周 pilot，8–12 周全量
+
+**Pilot（2–4 周）：** 10–15 个深测项目，先跑 C1/C2/C6 + C3 host install + 1 个市场差异表；不做全量修复、不做性能排名。
+
+**Full（8–12 周）：** 30 个项目、hold-out、性能子集、2–3 个伙伴 review。
+
+| 周 | 动作 | Gate |
+| --- | --- | --- |
+| 0–1 | 冻结 checklist；预注册 10 项目；golden set + 负向 control；host harness（hyzyn live-host-smoke 或 dsh-clinic D + 真 host） | 工具在 golden set 上 FP/FN ≤10%；C3 可跑；skip/degraded 不当 pass |
+| 2–3 | 30 项目静态 baseline；10–15 个 host install；市场差异表 | 至少 10 个 C3 有结果；至少 2 家市场规则映射 |
+| 4–5 | 10–15 个本地修复实验；before/after；performance 子集 | 0 第三方公开改动；每个修改只动一个变量 |
+| 6–7 | BOOST B1–B6；hold-out 验证；写 aggregate | hold-out 不调参；性能波动 >20% 标 not-run |
+| 8–10 | 2–3 个伙伴私下预览；根据反馈修标准；公开匿名 aggregate | 2 家愿意 review；0 点名/排名；官方知情 |
+| 11–12 | 决定下一轮：council/更多市场/BOOST B8 | 2 家公开采用前不叫 consensus/standard |
+
+## 8. 风险与 go/no-go
+
+- 样本偏差/过拟合 → 预注册、分层、30% hold-out、冻结规则。
+- 工具偏差 → golden set、负向 control、FP/FN 分开报；工具 bug 修工具，不修第三方。
+- 第三方/许可 → 本地副本、匿名、书面许可；AGPL/NC-SA 不混。
+- 性能噪声 → 固定环境、n≥20、raw data、波动 >20% not-run。
+- 市场对比 → 差异表，不打分；市场规则权威。
+- 沟通 → 先私聊预览、opt-in、不公开失败名单；官方/群先知情。
+- 单人时间 → 8–12 周 pilot；不做“全量修所有项目”。
+
+**Go/No-go：**
+1. golden set FP/FN ≤10%，否则先修工具；
+2. 至少 10 个 host install 有结果，否则不叫“全量”；
+2b. C3/C4/C5/C6 是否降为 FLOW/profile 必须有数据；不要先改代码再找证据；
+3. 第三方具名/PR 有书面许可，否则只本地匿名；
+4. 性能固定环境 n≥20，否则 B8 not-run；
+5. 2 家市场愿意私下 review，否则只发匿名方法/聚合。
+
+## 9. 现在先做 3 件
+
+1. **冻结 + 预注册**：10 个项目、假设、hold-out；不先改规则。
+2. **C3 host harness**：临时 `DSH_HOME`/profile，安装→启动→禁用→卸载；先在 10 个项目上跑；同时记录 C3 是否只是 FLOW。
+3. **一张市场差异表 + 一个伙伴预览**：优先 awesome 或 dsh-market；只私下给，不公开；用“你们的规则是权威”开场。
+
+## 10. 来源
+
+- `omdsh-dev/plugin-template`（README、docs/dsh-plugin-contracts.md、tests/harness.ts、tests/plugin.test.ts）
+- `morlay/dsh-plugin`（devpackages/tsdown-plugin-cordis、devkit、dshdev、dts-prepare）
+- `hyzyn/dsh-plugin-kit`（live-host-smoke、check-package-contents、check-dsh-peers）、`dsh-tauri`/`dsh-web`/`dsh-TUI`/`DSH-better-sidebar` 的 tests/scripts
+- `PerryLink/dsh-primer`（guide）与 `dsh-clinic`（doctor）SPEC；各市场 README/CONTRIBUTING/CI；dshfind/dshbase/dsh-plugin.market 的标签/数据
+- 本仓库：`adoption/entry/real-scan-2026-11-02.json`、`reviews/2026-11-02-p0-self-review.md`

package/docs/incident.md

@@ -0,0 +1,37 @@
+# Incident runbook
+
+> Scope: evidence errors, unauthorized merges, leaked secrets, and a false label.
+> This is not a security incident response service for third-party plugins.
+
+## 1. Detect
+
+- CI failure, appeal, user report, surveillance mismatch, or a maintainer notice.
+
+## 2. Contain
+
+- If a label is disputed: set `under-review`, hide the old PASS, keep history.
+- If a secret leaked: revoke the token first, then rotate; do not paste it.
+- If an AI merged outside its allowlist: revert, disable the token, record.
+
+## 3. Assess
+
+- Which record/subject/version? Which clause? Is it factual, procedural, or COI?
+- Does it affect a third party? If yes, notify via the public record.
+
+## 4. Correct
+
+- Issue a **new** record; do not edit the old PASS into a FAIL.
+- Append a decision-log entry and (if needed) a dispute JSON.
+- If the error was ours, say so; do not blame the plugin.
+
+## 5. Learn
+
+- Add a regression test/fixture.
+- Update the checklist or the taxonomy.
+- Record the time to detect and time to correct.
+
+## Red lines
+
+- Never delete a dispute or audit record.
+- Never restore a stale/withdrawn label to “valid”.
+- Never claim “no impact” without checking.

package/docs/onboarding.md

@@ -0,0 +1,27 @@
+# Onboarding / offboarding checklist
+
+> Access is granted per role, per scope, for a fixed term. No shared passwords.
+
+## Onboarding
+
+- [ ] Read `GOVERNANCE.md`, `44`, `governance/contribution-terms.md`.
+- [ ] Sign DCO + DCT; `ai-clerk` adds `contributors.json`.
+- [ ] Disclose employer/funding/own plugins in `coi.json`.
+- [ ] Enable 2FA; use your own token (never a shared one).
+- [ ] Role recorded in `roles.json`/`reviewers.json` with `expiresAt`.
+- [ ] For maintainers: one supervised release before write access.
+- [ ] For reviewers: 10 known answers + 2 shadow reviews.
+
+## Offboarding
+
+- [ ] Remove write access; keep the historical review/decision records.
+- [ ] Set role `dormant`/`emeritus`; do not delete contributions.
+- [ ] Rotate any token the person could read.
+- [ ] Record the change in `decision-log.md`.
+- [ ] If the organizer leaves: follow `CONTINUITY.md`; no AI inherits ownership.
+
+## Access inventory
+
+Keep this in a private file, never in git: GitHub, npm, PyPI, CI secrets, domain,
+and recovery codes. The public repo only records **who has which role**, not the
+secrets.

package/docs/privacy.md

@@ -0,0 +1,12 @@
+# Privacy and data handling
+
+- Do not commit tokens, session data, private plugin code, or personal data.
+- DCO `Signed-off-by` includes a name/email in public git history; use a
+  dedicated email if needed.
+- `contributors.json` records handle, level, and COI summary only. No address,
+  phone, ID number, or private employment details.
+- Evidence may include public repo URLs; do not add private dashboards.
+- A takedown request for personal data: remove it from the current tree and
+  record the action; git history may still contain it (say so honestly).
+- This project is not a data controller for third-party plugins; it only stores
+  what contributors publish.

package/docs/release.md

@@ -0,0 +1,16 @@
+# Release checklist
+
+1. `node --test 40-checker/test` — 0 fail.
+2. `governance-check`, `attestation-check`, `claim-check`, `clause-map`,
+   `evidence-chain verify`, `stale-check` — 0 error.
+3. Regenerate `dashboard.json` and the attestation index; do not hand-edit a PASS.
+4. Update `CHANGELOG.md` and the version in `spec/levels.json`/`scheme.json`.
+5. Tag `vX.Y.Z`; push; wait for `governance` CI.
+6. Publish npm/PyPI only after the tag; verify with a registry read.
+7. Record the result in `governance/decision-log.md`.
+
+## Rollback
+
+- Revert the tag/commit; keep the evidence.
+- Mark affected attestations `stale` or `withdrawn`; never delete them.
+- If a publish is bad, deprecate the version; do not reuse the version number.

package/docs/scope-security.md

@@ -0,0 +1,62 @@
+# Scope and security boundary (per level)
+
+> **Not a security audit.** This page exists so a reader cannot mistake a level
+> record for a safety, privacy, or production claim.
+
+## ENTRY
+
+**What it covers:** manifest/patch parse (G0), compose/row resolution (G1), host
+compatibility input (G2), real boot to ACTIVE/disabled (G3), client static shape
+(G4), freshness anchor (G5).
+
+**What it does not cover:**
+
+- vulnerability scanning, dependency CVE status, runtime isolation;
+- UI rendering or interaction;
+- data handling/privacy;
+- production load, cost, or provider behavior;
+- “safe to install”.
+
+**Security is not an ENTRY gate.** Security/brand findings are WARN-only in the
+gate; they are escalated to FLOW/governance, not used to fail ENTRY.
+
+## FLOW
+
+**What it covers:** install/remove (J1), enable/disable (J2), configuration (J3),
+first success (J4), permissions/disclosure (J5), upgrade/rollback/uninstall (J6),
+observability (J7), client/server shapes (J8).
+
+**What it does not cover:**
+
+- whether a disclosed permission is ethically acceptable;
+- secrets management beyond “no credential in the repo”;
+- multi-user isolation;
+- accessibility, i18n, visual design;
+- every platform combination.
+
+J5 is a **disclosure** check, not a security review.
+
+## BOOST
+
+**What it covers:** effect cleanup, inject/wait, volatile config, event modes,
+seam/isolate/group, prefix stability, bounded parallelism/jobs, composition.
+
+**What it does not cover:**
+
+- production latency/cost (unless a real provider baseline is recorded);
+- correctness of the feature;
+- security of the optimization;
+- “faster” as a general claim.
+
+## All levels
+
+- no official endorsement, no accreditation, no certification body;
+- no legal warranty, no insurance, no damages;
+- evidence expires; `stale` is not valid;
+- a `PASS` without a command/artifact is a bug.
+
+## How to report a security issue
+
+Do **not** open a public issue with exploit details. Use the repository’s
+private security channel (`SECURITY.md`). This project is not a CVE authority
+and cannot coordinate disclosure for third-party plugins.

package/docs/status-taxonomy.md

@@ -0,0 +1,54 @@
+# Status and failure taxonomy
+
+> The most common way a record lies is not a wrong number; it is a wrong status.
+> This page fixes the vocabulary.
+
+## Record status (per level)
+
+| Status | Meaning | Is it a pass? |
+|---|---|---|
+| `recognized` | static recognition only | no runtime claim |
+| `static` | static checks ran | no runtime claim |
+| `runtime` | a real boot/journey ran | yes for that scope |
+| `manual` | a human ran the journey and recorded it | yes for FLOW only |
+| `measured` | a benchmark was run and recorded | yes for BOOST only |
+| `blocked` | environment/host/service missing | **no** |
+| `failed` | the subject failed a check | **no** |
+| `not-run` | the check did not run | **no** |
+| `not-measured` | no BOOST evidence | **no** |
+| `not-applicable` | the check does not apply (data/docs repo) | **no** |
+| `stale` | evidence expired or host anchor changed | **no** |
+| `withdrawn` | author/project withdrew it | **no** |
+
+## Rules
+
+1. `blocked` ≠ `failed`: infrastructure/host is not the plugin’s fault.
+2. `not-run` ≠ `not-applicable`: “did not run” and “does not apply” are different.
+3. `static` ≠ `runtime`; `client-module-loaded` ≠ UI rendered.
+4. `measured` only for BOOST; `manual` only for FLOW.
+5. `stale` is computed by the reader from `expiresAt`, not trusted from the writer.
+6. A `PASS` needs a command/artifact; a skip must never be upgraded.
+
+## Failure classes
+
+| Class | Example | Who fixes |
+|---|---|---|
+| `subject-fail` | G1 row unresolved, G4 export missing | plugin author |
+| `host-blocked` | host gate rejects, profile service missing | host/adapter |
+| `fixture-fail` | fixture cannot boot; sample is wrong | standards maintainer |
+| `tool-error` | checker crash, JSON parse error | checker maintainer |
+| `stale` | expiry passed | author re-runs |
+| `withdrawn` | evidence invalid/misused | organizer |
+
+The checker must not map `host-blocked` or `tool-error` to `subject-fail`.
+
+## Reporting template
+
+```text
+status: <one of the above>
+class: subject-fail | host-blocked | fixture-fail | tool-error | stale | withdrawn
+evidence: <command + artifact>
+scope: dsh=<v> cordis=<v> platform=<v>
+expiresAt: <date>
+limitation: <one sentence>
+```

package/docs/sunset.md

@@ -0,0 +1,44 @@
+# Level sunset and retirement
+
+> A level is a **draft practice**, not a permanent standard. It can be frozen,
+> superseded, or retired. Retirement is a normal outcome, not a failure.
+
+## States
+
+| State | Meaning | Display |
+|---|---|---|
+| `draft` | current, provisional | label allowed with `draft` |
+| `frozen` | no new clauses; existing evidence stays valid until expiry | label + `frozen` |
+| `deprecated` | superseded; new claims discouraged | label + `deprecated`; no new attestation |
+| `retired` | no longer maintained | historical only; `stale` |
+
+## Triggers
+
+- host minor changes make a gate unjudgeable;
+- a clause is proven wrong by a counterexample;
+- no adopter for 12 months and no maintainer;
+- a replacement level is adopted;
+- the project loses its maintainer (bus factor event).
+
+## Procedure
+
+1. `ai-clerk` opens a sunset proposal with the clause IDs and evidence.
+2. 7-day public comment (Class B); record objections.
+3. Organizer decides: `frozen` / `deprecated` / `retired`.
+4. Existing attestations are not deleted; they are marked `stale` or
+   `superseded` at their expiry.
+5. New claims stop on the decision date; a later claim must use the replacement.
+
+## What must not happen
+
+- silently deleting a level or its evidence;
+- keeping a `retired` label in the “valid” filter;
+- converting old grades/levels to a new scheme without a new record;
+- claiming “consensus” for a solo decision.
+
+## Machine support
+
+- `spec/levels.json` carries `review` and `expiryDays`;
+- `scripts/stale-check.mjs` recomputes expiry;
+- `governance/decision-log.md` records the sunset decision;
+- `publish/attestations/index.json` keeps the historical record.

package/docs/translating.md

@@ -0,0 +1,12 @@
+# Translation rules
+
+> Translations are derived text; the Chinese source is normative until the
+> organizer says otherwise.
+
+1. Keep clause IDs, gate IDs, and status words unchanged.
+2. Do not translate `not-run`, `blocked`, `measured`, `manual`, `stale` into
+   words that imply pass/fail.
+3. Add `Translation-Of: <commit>` to the commit trailer.
+4. One PR per language; do not bundle a wording change with a translation.
+5. A translation that changes meaning is a bug; link the source line.
+6. Machine translation is allowed with human review; disclose it.

package/package.json

@@ -1,26 +1,57 @@
 {
-  "version": "0.0.2",
-  "description": "Name reservation + pointer for the DSH plugin standards draft (non-official).",
-  "license": "MIT",
-  "repository": {
-    "type": "git",
-    "url": "https://github.com/PerryLink/dsh-standards.git"
+  "name": "dsh-standards",
+  "version": "1.0.0",
+  "description": "DSH plugin standards and conformance checks: ENTRY/FLOW checks plus BOOST guidance. Non-official; no certification or ranking.",
+  "type": "module",
+  "main": "40-checker/src/index.mjs",
+  "exports": {
+    ".": "./40-checker/src/index.mjs",
+    "./package.json": "./package.json"
   },
-  "homepage": "https://github.com/PerryLink/dsh-standards",
-  "keywords": [
-    "dsh",
-    "deepseek-harness",
-    "plugin",
-    "standards"
-  ],
-  "main": "index.mjs",
   "bin": {
-    "dsh-standards": "bin/dsh-standards.mjs"
+    "dsh-standards": "./bin/dsh-standards.mjs",
+    "dsh-levels": "./bin/dsh-levels.mjs"
   },
   "files": [
-    "index.mjs",
+    "40-checker",
+    "spec",
+    "docs",
     "bin",
-    "README.md"
+    "README.md",
+    "LICENSE",
+    "LICENSE-APACHE",
+    "LICENSE-CC-BY",
+    "NOTICE.md"
+  ],
+  "scripts": {
+    "test": "node --test 40-checker/test/*.test.mjs",
+    "test:fixtures": "node bin/dsh-standards.mjs 40-checker/fixtures/good --scope entry --entry-claim static --format text"
+  },
+  "engines": {
+    "node": ">=20"
+  },
+  "keywords": [
+    "dsh",
+    "dsh-plugin",
+    "deepseek-harness",
+    "cordis",
+    "standards",
+    "checker",
+    "conformance",
+    "entry",
+    "flow",
+    "boost"
   ],
-  "name": "dsh-standards"
+  "license": "Apache-2.0",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/PerryLink/dsh-standards.git"
+  },
+  "homepage": "https://github.com/PerryLink/dsh-standards#readme",
+  "bugs": {
+    "url": "https://github.com/PerryLink/dsh-standards/issues"
+  },
+  "publishConfig": {
+    "access": "public"
+  }
 }

package/spec/README.md

@@ -0,0 +1,27 @@
+# dsh-standards specification tree
+
+One project, three adoption levels. Each level has its own maturity; they share
+the glossary, evidence anchor, and checker.
+
+| Level | Document | Maturity | Evidence |
+|---|---|---|---|
+| ENTRY | [../31-DSH-ENTRY入门规范.md](../31-DSH-ENTRY入门规范.md) | public-preview | 18/18 recognized; 10/18 base G3, 8 clean after gate; client 15/15 loaded |
+| FLOW | [../32-DSH-FLOW顺畅规范.md](../32-DSH-FLOW顺畅规范.md) | draft | J1/J2/J4/J6 runtime on fixture; 5 real packages; J3/J5/J7/J8 not-run |
+| BOOST | [../33-DSH-BOOST加速实践.md](../33-DSH-BOOST加速实践.md) | draft | LEV-1–8 8/8 measured at synthetic/loader/provider-capture level |
+
+Machine-readable:
+
+- [levels.json](levels.json) — levels, gates, evidence states, crosswalk.
+- [clauses.json](clauses.json) — stable clause IDs mapped to gates (`clause-map.mjs`).
+- [reference-suite.json](reference-suite.json) — versioned samples, anchors, change rules.
+- [schemas/evidence.schema.json](schemas/evidence.schema.json) — `dsh-levels/1` record.
+- [schemas/attestation.schema.json](schemas/attestation.schema.json) — `dsh-attestation/1` quality record.
+- [schemas/report.schema.json](schemas/report.schema.json) — `dsh-report/1` inspection report (no grade).
+- [../40-checker/scripts/level-audit.mjs](../40-checker/scripts/level-audit.mjs) — one-command audit.
+- [../publish/level-labels.example.json](../publish/level-labels.example.json) — market labels.
+
+Rules:
+
+- Levels depend on `common` only; do not import each other.
+- A level is not a certification, badge, grade, or score.
+- `dsh-standard` is an alias; `dsh-standards` is canonical.

package/spec/boost/README.md

@@ -0,0 +1,8 @@
+# BOOST
+
+Leverage surface: Cordis-native practices with evidence, not a grade.
+
+- Practice text: [../../33-DSH-BOOST加速实践.md](../../33-DSH-BOOST加速实践.md)
+- Deep research: [../../91-增效实践内核研究.md](../../91-增效实践内核研究.md)
+- Reference: `examples/boost/lev7-bounded-batch.mjs`; benchmark: `examples/boost/results/lev7.json`
+- Current: LEV-3 loader-level 0/20 vs 20/20 remount; LEV-6 provider-capture 100% vs 0% prompt reuse; LEV-7 8.24× synthetic. None is a production baseline.