dsh-data-domain-knowledge 0.1.0-alpha.1

package/lib/types/config-schema.d.ts

@@ -0,0 +1,35 @@
+/**
+ * 插件 entry 的配置 schema（0.1.7）。
+ *
+ * ## 为什么单独一个模块
+ *
+ * 0.1.7 删掉了「设置命名空间 + `settings.installSection`」这一整套 API，改成
+ * **设置就是插件自己 entry 的 config**：schema 在这里声明，初值由 profile patch 的
+ * `config:` 提供，设置页里的改动由 `configEditor` 写回同一处，客户端按**路径**寻址
+ * （`{op:'set', path:['intent','categories']}`）。
+ *
+ * 三组沿用原来的三个命名空间边界，因为它们在语义上确实是三块（知识库 / 意图引擎 /
+ * 转换内核），缓存键与默认值各不相同 —— 变的只是「它们住在同一个 entry 下」。
+ *
+ * 「单独一个模块」的理由和 `settings-schema.ts`、`intent/settings-schema.ts` 一样，
+ * 是**可测**：`tests/settings-schema.test.mjs` 要能 import 这个 schema 并把它解析
+ * 一遍，而 `index.ts` 会拉进工具层（`dsh-tools` 的 `dsh-sandbox` peer 没有安装，
+ * `node --test` 里拉不起来）。所以组合必须发生在没有工具依赖的地方。
+ *
+ * @module
+ */
+import type { KnowledgeConfig } from './types.js';
+import type { IntentSettings } from './intent/types.js';
+import type { ConvertPluginConfig } from './convert/types.js';
+/**
+ * 标注 `Schemastery` 是 TS2742 的要求（见 `intent/settings-schema.ts`）：导出物必须
+ * 有一个能被命名的类型，否则 `.d.ts` 里会写进 cosmokit 的 `.pnpm` 路径。
+ */
+export declare const Config: Schemastery;
+/** `Config` 的取值形态。三组都可选：缺的字段由 schema 默认值补齐。 */
+export interface KnowledgePluginConfig {
+    knowledge?: Partial<KnowledgeConfig>;
+    intent?: Partial<IntentSettings>;
+    convert?: Partial<ConvertPluginConfig>;
+}
+//# sourceMappingURL=config-schema.d.ts.map

package/lib/types/config-schema.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"config-schema.d.ts","sourceRoot":"","sources":["../../src/config-schema.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAMH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAA;AACjD,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAA;AACvD,OAAO,KAAK,EAAE,mBAAmB,EAAE,MAAM,oBAAoB,CAAA;AAE7D;;;GAGG;AACH,eAAO,MAAM,MAAM,EAAE,WAInB,CAAA;AAEF,+CAA+C;AAC/C,MAAM,WAAW,qBAAqB;IACpC,SAAS,CAAC,EAAE,OAAO,CAAC,eAAe,CAAC,CAAA;IACpC,MAAM,CAAC,EAAE,OAAO,CAAC,cAAc,CAAC,CAAA;IAChC,OAAO,CAAC,EAAE,OAAO,CAAC,mBAAmB,CAAC,CAAA;CACvC"}

package/lib/types/config-schema.js

@@ -0,0 +1,34 @@
+/**
+ * 插件 entry 的配置 schema（0.1.7）。
+ *
+ * ## 为什么单独一个模块
+ *
+ * 0.1.7 删掉了「设置命名空间 + `settings.installSection`」这一整套 API，改成
+ * **设置就是插件自己 entry 的 config**：schema 在这里声明，初值由 profile patch 的
+ * `config:` 提供，设置页里的改动由 `configEditor` 写回同一处，客户端按**路径**寻址
+ * （`{op:'set', path:['intent','categories']}`）。
+ *
+ * 三组沿用原来的三个命名空间边界，因为它们在语义上确实是三块（知识库 / 意图引擎 /
+ * 转换内核），缓存键与默认值各不相同 —— 变的只是「它们住在同一个 entry 下」。
+ *
+ * 「单独一个模块」的理由和 `settings-schema.ts`、`intent/settings-schema.ts` 一样，
+ * 是**可测**：`tests/settings-schema.test.mjs` 要能 import 这个 schema 并把它解析
+ * 一遍，而 `index.ts` 会拉进工具层（`dsh-tools` 的 `dsh-sandbox` peer 没有安装，
+ * `node --test` 里拉不起来）。所以组合必须发生在没有工具依赖的地方。
+ *
+ * @module
+ */
+import z from '@deepseek-ai/schemastery';
+import { KNOWLEDGE_SETTINGS_SCHEMA } from './settings-schema.js';
+import { INTENT_SETTINGS_SCHEMA } from './intent/settings-schema.js';
+import { CONVERT_SETTINGS_SCHEMA } from './convert/settings-schema.js';
+/**
+ * 标注 `Schemastery` 是 TS2742 的要求（见 `intent/settings-schema.ts`）：导出物必须
+ * 有一个能被命名的类型，否则 `.d.ts` 里会写进 cosmokit 的 `.pnpm` 路径。
+ */
+export const Config = z.object({
+    knowledge: KNOWLEDGE_SETTINGS_SCHEMA,
+    intent: INTENT_SETTINGS_SCHEMA,
+    convert: CONVERT_SETTINGS_SCHEMA,
+});
+//# sourceMappingURL=config-schema.js.map

package/lib/types/config-schema.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"config-schema.js","sourceRoot":"","sources":["../../src/config-schema.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,CAAC,MAAM,0BAA0B,CAAA;AACxC,OAAO,EAAE,yBAAyB,EAAE,MAAM,sBAAsB,CAAA;AAChE,OAAO,EAAE,sBAAsB,EAAE,MAAM,6BAA6B,CAAA;AACpE,OAAO,EAAE,uBAAuB,EAAE,MAAM,8BAA8B,CAAA;AAKtE;;;GAGG;AACH,MAAM,CAAC,MAAM,MAAM,GAAgB,CAAC,CAAC,MAAM,CAAC;IAC1C,SAAS,EAAE,yBAAyB;IACpC,MAAM,EAAE,sBAAsB;IAC9B,OAAO,EAAE,uBAAuB;CACjC,CAAC,CAAA"}

package/lib/types/convert/core-loader.d.ts

@@ -0,0 +1,32 @@
+/**
+ * Lazy, optional access to the conversion kernel.
+ *
+ * `dsh-convert-core` is declared as an **optionalDependency**: it
+ * pulls in `@firecrawl/anydoc`, which ships a platform-specific prebuilt native
+ * binding. A user who only wants to ingest Markdown must still be able to
+ * install and run the knowledge plugin, so the kernel is imported on first use
+ * and a missing kernel degrades to one actionable error instead of taking the
+ * whole plugin (SourceRaw, compile, recall) down with it.
+ *
+ * Guardrail this file exists to enforce: nothing else in the plugin may import
+ * the kernel eagerly.
+ */
+export type ConvertCoreModule = typeof import('dsh-convert-core');
+export type ConvertCoreLoader = () => Promise<ConvertCoreModule>;
+/** Raised when the optional kernel cannot be loaded. */
+export declare class ConvertCoreMissingError extends Error {
+    readonly code = "DEPENDENCY_MISSING";
+    readonly retryable = true;
+    readonly suggestion = "\u8F6C\u6362\u5185\u6838\u662F\u53EF\u9009\u4F9D\u8D56\uFF1A\u88C5\u4E0A dsh-convert-core \u540E\u91CD\u8BD5\uFF08pnpm --filter dsh-data-domain-knowledge install\uFF09\u3002knowledge \u7684\u5F55\u5165\u3001\u7F16\u8BD1\u4E0E\u53EC\u56DE\u4E0D\u4F9D\u8D56\u5B83\u3002";
+    constructor(cause?: unknown);
+}
+/** Load (and memoise) the kernel, or throw `ConvertCoreMissingError`. */
+export declare function loadConvertCore(): Promise<ConvertCoreModule>;
+/** Whether the kernel is already loaded; used by capabilities reporting. */
+export declare function convertCoreLoaded(): boolean;
+/**
+ * Test seam. Pass a loader to simulate a missing/broken kernel, or `undefined`
+ * to restore the real dynamic import. Never used by production code.
+ */
+export declare function setConvertCoreLoader(next?: ConvertCoreLoader): void;
+//# sourceMappingURL=core-loader.d.ts.map

package/lib/types/convert/core-loader.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"core-loader.d.ts","sourceRoot":"","sources":["../../../src/convert/core-loader.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,MAAM,MAAM,iBAAiB,GAAG,cAAc,kBAAkB,CAAC,CAAA;AAEjE,MAAM,MAAM,iBAAiB,GAAG,MAAM,OAAO,CAAC,iBAAiB,CAAC,CAAA;AAIhE,wDAAwD;AACxD,qBAAa,uBAAwB,SAAQ,KAAK;IAChD,QAAQ,CAAC,IAAI,wBAAuB;IACpC,QAAQ,CAAC,SAAS,QAAO;IACzB,QAAQ,CAAC,UAAU,iRAAe;gBACtB,KAAK,CAAC,EAAE,OAAO;CAK5B;AAUD,yEAAyE;AACzE,wBAAsB,eAAe,IAAI,OAAO,CAAC,iBAAiB,CAAC,CAWlE;AAED,4EAA4E;AAC5E,wBAAgB,iBAAiB,IAAI,OAAO,CAE3C;AAED;;;GAGG;AACH,wBAAgB,oBAAoB,CAAC,IAAI,CAAC,EAAE,iBAAiB,GAAG,IAAI,CAInE"}

package/lib/types/convert/core-loader.js

@@ -0,0 +1,60 @@
+/**
+ * Lazy, optional access to the conversion kernel.
+ *
+ * `dsh-convert-core` is declared as an **optionalDependency**: it
+ * pulls in `@firecrawl/anydoc`, which ships a platform-specific prebuilt native
+ * binding. A user who only wants to ingest Markdown must still be able to
+ * install and run the knowledge plugin, so the kernel is imported on first use
+ * and a missing kernel degrades to one actionable error instead of taking the
+ * whole plugin (SourceRaw, compile, recall) down with it.
+ *
+ * Guardrail this file exists to enforce: nothing else in the plugin may import
+ * the kernel eagerly.
+ */
+const INSTALL_HINT = '转换内核是可选依赖：装上 dsh-convert-core 后重试（pnpm --filter dsh-data-domain-knowledge install）。knowledge 的录入、编译与召回不依赖它。';
+/** Raised when the optional kernel cannot be loaded. */
+export class ConvertCoreMissingError extends Error {
+    code = 'DEPENDENCY_MISSING';
+    retryable = true;
+    suggestion = INSTALL_HINT;
+    constructor(cause) {
+        const detail = cause instanceof Error ? cause.message : cause === undefined ? '' : String(cause);
+        super(`转换内核 dsh-convert-core 不可用${detail ? `：${detail}` : ''}`);
+        this.name = 'ConvertCoreMissingError';
+    }
+}
+async function importKernel() {
+    return await import('dsh-convert-core');
+}
+let loader = importKernel;
+let cached;
+let inflight;
+/** Load (and memoise) the kernel, or throw `ConvertCoreMissingError`. */
+export async function loadConvertCore() {
+    if (cached)
+        return cached;
+    inflight ??= loader();
+    try {
+        const module = await inflight;
+        cached = module;
+        return module;
+    }
+    catch (cause) {
+        inflight = undefined;
+        throw cause instanceof ConvertCoreMissingError ? cause : new ConvertCoreMissingError(cause);
+    }
+}
+/** Whether the kernel is already loaded; used by capabilities reporting. */
+export function convertCoreLoaded() {
+    return cached !== undefined;
+}
+/**
+ * Test seam. Pass a loader to simulate a missing/broken kernel, or `undefined`
+ * to restore the real dynamic import. Never used by production code.
+ */
+export function setConvertCoreLoader(next) {
+    loader = next ?? importKernel;
+    cached = undefined;
+    inflight = undefined;
+}
+//# sourceMappingURL=core-loader.js.map

package/lib/types/convert/core-loader.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"core-loader.js","sourceRoot":"","sources":["../../../src/convert/core-loader.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAMH,MAAM,YAAY,GAAG,6GAA6G,CAAA;AAElI,wDAAwD;AACxD,MAAM,OAAO,uBAAwB,SAAQ,KAAK;IACvC,IAAI,GAAG,oBAAoB,CAAA;IAC3B,SAAS,GAAG,IAAI,CAAA;IAChB,UAAU,GAAG,YAAY,CAAA;IAClC,YAAY,KAAe;QACzB,MAAM,MAAM,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAA;QAChG,KAAK,CAAC,4BAA4B,MAAM,CAAC,CAAC,CAAC,IAAI,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAA;QAC/D,IAAI,CAAC,IAAI,GAAG,yBAAyB,CAAA;IACvC,CAAC;CACF;AAED,KAAK,UAAU,YAAY;IACzB,OAAO,MAAM,MAAM,CAAC,kBAAkB,CAAC,CAAA;AACzC,CAAC;AAED,IAAI,MAAM,GAAsB,YAAY,CAAA;AAC5C,IAAI,MAAqC,CAAA;AACzC,IAAI,QAAgD,CAAA;AAEpD,yEAAyE;AACzE,MAAM,CAAC,KAAK,UAAU,eAAe;IACnC,IAAI,MAAM;QAAE,OAAO,MAAM,CAAA;IACzB,QAAQ,KAAK,MAAM,EAAE,CAAA;IACrB,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAA;QAC7B,MAAM,GAAG,MAAM,CAAA;QACf,OAAO,MAAM,CAAA;IACf,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,QAAQ,GAAG,SAAS,CAAA;QACpB,MAAM,KAAK,YAAY,uBAAuB,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,uBAAuB,CAAC,KAAK,CAAC,CAAA;IAC7F,CAAC;AACH,CAAC;AAED,4EAA4E;AAC5E,MAAM,UAAU,iBAAiB;IAC/B,OAAO,MAAM,KAAK,SAAS,CAAA;AAC7B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,oBAAoB,CAAC,IAAwB;IAC3D,MAAM,GAAG,IAAI,IAAI,YAAY,CAAA;IAC7B,MAAM,GAAG,SAAS,CAAA;IAClB,QAAQ,GAAG,SAAS,CAAA;AACtB,CAAC"}

package/lib/types/convert/host.d.ts

@@ -0,0 +1,88 @@
+/**
+ * The convert host — the DSH-facing half of multi-format conversion.
+ *
+ * Since 0018 this lives inside `plugins/dsh-data-domain-knowledge` (issue 0018):
+ * conversion is a sub-capability of knowledge, mounted as one plugin, with one
+ * sidebar entry and one settings surface for the user. The kernel
+ * (`dsh-convert-core`) stays an independent, optional package that
+ * `core-loader.ts` imports lazily.
+ *
+ * Layering this file must keep:
+ *
+ *   kernel (convert-core)   input → knowledge-ready Markdown + assets + console
+ *   this host               settings, agent tools, /convert route, ingestion
+ *
+ * Nothing about parsing lives here; nothing about DSH lives in the kernel. That
+ * is why the same console still works standalone at `tita-convert`.
+ */
+import type { Context } from '@deepseek-ai/cordis';
+import { type IngestTarget } from './ingest.js';
+import { type ConvertPluginConfig } from './types.js';
+export { CONVERT_SETTINGS_NAMESPACE, DEFAULT_CONVERT_CONFIG, defaultConvertRoot } from './types.js';
+/** Route prefix the console and API are mounted on. Frozen contract. */
+export declare const MOUNT_PATH = "/convert";
+/**
+ * 转换的设置 schema。
+ *
+ * 0029 起它**没有路径字段**：交付物与工作区都从知识插件的 `agentRoot` 推出来
+ * （`<agentRoot>/output/<日期>/`、`<agentRoot>/tmp/`）。一个 Agent 只有一个根路径，
+ * 所以这里再放一个 `rootDir` 就等于把「东西存在哪」这个答案劈成两半。
+ *
+ * 定义本身搬到了 `./settings-schema.js`（纯数据、可被 `Config` 组合并在
+ * `node --test` 里独立解析）；这里 re-export，保持既有出口不变。
+ */
+export { CONVERT_SETTINGS_SCHEMA } from './settings-schema.js';
+/** What the convert host needs from the knowledge plugin it now lives in. */
+export interface ConvertHostDeps {
+    /** The live knowledge runtime; the seam may only call `addSource` on it. */
+    runtime(): IngestTarget;
+    /** Absolute knowledge root the runtime currently writes to. */
+    knowledgeRoot(): string;
+    /**
+     * 当前工作项目（交付物的落点）。省略时退回进程的工作目录。
+     *
+     * 转换内核是**宿主级单例**，而「工作项目」是**每个会话各自**的 —— 内核这一层拿不到
+     * 会话，所以这里只能给一个宿主级的答案。单项目部署（一个 DSH 进程服务同一个项目）下
+     * 两者一致；多项目共用同一个宿主时，交付物会集中落到宿主启动目录下，而不是各自项目里。
+     * 要做到完全按会话分，需要内核支持**按请求**解析输出目录，那是内核侧的一处改造。
+     */
+    workspace?(): string;
+    /**
+     * Resolve an ingest target for one knowledge base (issue 0019). Optional so the
+     * host still works when the registry is unavailable; when it is absent every
+     * conversion lands in the default knowledge base.
+     */
+    resolveTarget?(kbId?: string): {
+        runtime: IngestTarget;
+        knowledgeRoot: string;
+    };
+    /**
+     * Tita Agent 根路径（0029）。转换的交付物与工作区都从它推出来，所以转换
+     * **没有自己的路径设置** —— 一个 Agent 只有一个根路径。
+     */
+    agentRoot(): string;
+    /**
+     * 已解析的 `convert` 设置组（来自插件 entry 的 config）。
+     *
+     * 0.1.7 起设置不再是命名空间，而是随 `apply(ctx, config)` 进来的 entry config，
+     * 所以这里由调用方把那一组传下来。省略时用 `DEFAULT_CONVERT_CONFIG`。
+     *
+     * 允许给**函数**的理由与意图那组相同（见 `intent/host.ts` 的 `IntentHostDeps.config`）：
+     * 字段是 volatile，解包是一次快照，而 volatile 改动不会重载插件。
+     */
+    config?: Partial<ConvertPluginConfig> | (() => Partial<ConvertPluginConfig>);
+}
+/**
+ * `applyConvert` 交给组合根的那一个把手，与 `intent/host.ts` 的 `IntentHost` 同形。
+ *
+ * 转换这一组也有 volatile 字段（`maxConcurrent` / `maxInputBytes` / `retentionDays` /
+ * `toolWaitMs` / `ingestEnabled`），其中一个解包漏了就是沉默的错误：`maxInputBytes`
+ * 是引用时 `info.size > config.maxInputBytes` 恒为 false，体积上限静默消失；
+ * `ingestEnabled` 是引用时它恒为真，开关关不掉。所以这一组也要能就地重灌。
+ */
+export interface ConvertHost {
+    /** 按**当前**的 entry config 重灌一次设置，并按新值重置转换内核。 */
+    reprocess(): void;
+}
+export declare function applyConvert(ctx: Context, deps: ConvertHostDeps): ConvertHost;
+//# sourceMappingURL=host.d.ts.map

package/lib/types/convert/host.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"host.d.ts","sourceRoot":"","sources":["../../../src/convert/host.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAKH,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,qBAAqB,CAAA;AAOlD,OAAO,EAAsC,KAAK,YAAY,EAAE,MAAM,aAAa,CAAA;AACnF,OAAO,EAML,KAAK,mBAAmB,EAEzB,MAAM,YAAY,CAAA;AAEnB,OAAO,EAAE,0BAA0B,EAAE,sBAAsB,EAAE,kBAAkB,EAAE,MAAM,YAAY,CAAA;AAEnG,wEAAwE;AACxE,eAAO,MAAM,UAAU,aAAa,CAAA;AAIpC;;;;;;;;;GASG;AACH,OAAO,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAA;AAE9D,6EAA6E;AAC7E,MAAM,WAAW,eAAe;IAC9B,4EAA4E;IAC5E,OAAO,IAAI,YAAY,CAAA;IACvB,+DAA+D;IAC/D,aAAa,IAAI,MAAM,CAAA;IACvB;;;;;;;OAOG;IACH,SAAS,CAAC,IAAI,MAAM,CAAA;IACpB;;;;OAIG;IACH,aAAa,CAAC,CAAC,IAAI,CAAC,EAAE,MAAM,GAAG;QAAE,OAAO,EAAE,YAAY,CAAC;QAAC,aAAa,EAAE,MAAM,CAAA;KAAE,CAAA;IAC/E;;;OAGG;IACH,SAAS,IAAI,MAAM,CAAA;IACnB;;;;;;;;OAQG;IACH,MAAM,CAAC,EAAE,OAAO,CAAC,mBAAmB,CAAC,GAAG,CAAC,MAAM,OAAO,CAAC,mBAAmB,CAAC,CAAC,CAAA;CAC7E;AAED;;;;;;;GAOG;AACH,MAAM,WAAW,WAAW;IAC1B,+CAA+C;IAC/C,SAAS,IAAI,IAAI,CAAA;CAClB;AAwFD,wBAAgB,YAAY,CAAC,GAAG,EAAE,OAAO,EAAE,IAAI,EAAE,eAAe,GAAG,WAAW,CA+X7E"}