@v587d/capital-generation 0.0.0-stage → 2.6.0

package/lib/data-collector/tools.js

@@ -0,0 +1,216 @@
+const jsonObject = (properties = {}, required = []) => ({ type: 'object', properties, required, additionalProperties: false });
+const freeObject = { type: 'object', additionalProperties: true };
+const MAX_ID_LENGTH = 256;
+const MAX_PARAMS_BYTES = 64 * 1024;
+function boundedString(value, name, maxLength) {
+    if (typeof value !== 'string' || value.length === 0)
+        throw new Error(`${name} is required`);
+    if (value.length > maxLength)
+        throw new Error(`${name} exceeds maximum length ${maxLength}`);
+    return value;
+}
+function boundedParams(value) {
+    if (value === undefined)
+        return {};
+    if (!value || typeof value !== 'object' || Array.isArray(value))
+        throw new Error('params must be an object');
+    let encoded;
+    try {
+        encoded = JSON.stringify(value);
+    }
+    catch {
+        throw new Error('params must be JSON-serializable');
+    }
+    if (encoded.length > MAX_PARAMS_BYTES)
+        throw new Error(`params exceeds maximum size ${MAX_PARAMS_BYTES} bytes`);
+    return value;
+}
+/**
+ * DatasetRef 输出结构（模型可见协议）：只有元数据。任何 raw rows、内部
+ * data_key、绝对路径字段都不允许出现。
+ */
+const datasetRefSchema = () => jsonObject({
+    dataset_id: { type: 'string' },
+    task_id: { oneOf: [{ type: 'string' }, { type: 'null' }] },
+    session_id: { type: 'string' },
+    artifact_ref: { type: 'string' },
+    format: { type: 'string' },
+    capability: { type: 'string' },
+    source_label: { type: 'string' },
+    schema: { oneOf: [{ type: 'object', additionalProperties: true }, { type: 'null' }] },
+    row_count: { oneOf: [{ type: 'integer' }, { type: 'null' }] },
+    captured_at: { type: 'integer' },
+    retention_until: { type: 'integer' },
+    params_digest: { type: 'string' },
+}, ['dataset_id', 'task_id', 'session_id', 'artifact_ref', 'format', 'capability', 'source_label', 'schema', 'row_count', 'captured_at', 'retention_until', 'params_digest']);
+function render(_args, value) {
+    // ToolOutputDefinition.render(args, value)：第一参是入参、第二参才是返回值。
+    return [{ type: 'text', text: JSON.stringify(value) }];
+}
+function tool(name, description, parameters, schema, execute) {
+    return { name, description, parameters, output: { schema, render }, execute };
+}
+/**
+ * 调用方 Agent session（官方注入，不可由模型伪造）。请求归属恒为当前调用者，
+ * 不存在"代理请求"概念。workspace cwd 由 session.header.cwd 提供。
+ */
+const callerSession = (exec) => {
+    const session = exec.agent?.session;
+    if (session)
+        return session;
+    throw new Error('this tool requires a calling agent session (exec.agent.session was undefined)');
+};
+const delegatedSession = (exec, toolName) => {
+    const session = callerSession(exec);
+    if (typeof session.header?.parentSession !== 'string' || session.header.parentSession.length === 0) {
+        throw new Error(`${toolName} is restricted to delegated data agents; the main Agent must delegate this request`);
+    }
+    return session;
+};
+/** dc_status 真实执行计数（进程内递增，返回给模型可见）。 */
+const dcStatusCalls = { current: 0 };
+function capabilityError(code, detail) {
+    const error = new Error(`${code}: ${detail}`);
+    error.code = code;
+    return error;
+}
+function errorCode(error) {
+    return error && typeof error === 'object' && typeof error.code === 'string'
+        ? error.code
+        : undefined;
+}
+/**
+ * 数据源在 normalizeParams 阶段抛出的错误对模型来说都是可修正的参数错误。
+ * 保留数据源的字段级原因，同时把恢复动作固定成一次 describe -> 修正 -> 重试，
+ * 避免子 Agent 只看到 "must be ..." 后凭记忆反复猜参数。
+ */
+function requestParamsError(error, capability) {
+    if (errorCode(error))
+        return undefined;
+    const message = error instanceof Error ? error.message : String(error);
+    if (!/(?:\bparameter\b|\bparams\b|missing required parameter|unsupported parameter|at least one of)/i.test(message))
+        return undefined;
+    const detail = `request_params_invalid: ${message}. 请先调用 describe_capability({ "capability": ${JSON.stringify(capability)} }) 重新读取该能力的 input_schema，再严格按 schema 修正 params 后重试 request_data；不要凭记忆猜测参数。 Please read describe_capability(${JSON.stringify(capability)}) before retrying request_data.`;
+    const failure = new Error(detail);
+    failure.code = 'request_params_invalid';
+    return failure;
+}
+function resolveCapability(hub, value) {
+    if (value === undefined || value === null || value === '') {
+        throw capabilityError('capability_required', '缺少 capability 参数。请先调用 list_capabilities() 取回能力目录，再调用 describe_capability({ "capability": "<目录中的名字>" }) 读取该能力契约。');
+    }
+    if (typeof value !== 'string') {
+        throw capabilityError('capability_invalid', `capability 必须是字符串（收到 ${Array.isArray(value) ? 'array' : typeof value}）。一次只描述一个能力；请从 list_capabilities() 的目录中原样复制名字，再调用 describe_capability。`);
+    }
+    const name = value.trim();
+    if (name.length === 0) {
+        throw capabilityError('capability_required', 'capability 是空白字符串。请先调用 list_capabilities()，再把目录中的 capability 名原样传给 describe_capability。');
+    }
+    if (name.length > MAX_ID_LENGTH) {
+        throw capabilityError('capability_invalid', `capability 超过 ${MAX_ID_LENGTH} 字符，不是合法能力名。请从 list_capabilities() 的目录里原样复制一个名字。`);
+    }
+    if (name.includes(',')) {
+        throw capabilityError('capability_invalid', `capability 不接受逗号分隔的多个名字（收到 "${name}"）。一次只调用 describe_capability 描述一个能力；需要多个就分多次调用。`);
+    }
+    if (!hub.describeCapability(name)) {
+        const available = hub.capabilityNames();
+        if (available.length === 0) {
+            throw capabilityError('capability_catalog_empty', '当前没有任何已注册的数据能力（数据源注册没有生效）。注意：缺凭据**不会**让目录变空——同花顺能力始终在目录里，只在取数那一刻点名 FUYAO_API_KEY 失败。本工具此刻对任何名字都会失败——不要反复重试，也不要编造能力名；请把该状态回告主 Agent，必要时用 dc_status 核对 registered_capabilities 与 api_key。');
+        }
+        throw capabilityError('capability_unknown', `未注册的能力 "${name}"。当前可用能力：${available.join(', ')}。请从 list_capabilities() 返回的目录里原样复制一个名字，再调用 describe_capability；不要自行编造、缩写或改写。`);
+    }
+    return name;
+}
+export function registerDataCollectorTools(ctx, hub, diagnostics) {
+    const tools = ctx.get('tools');
+    if (!tools)
+        return;
+    const registrations = [
+        tool('request_data', '向数据管道请求数据：宿主先按 capability+params 校验当前 session 的已验证 manifest；force_refresh=false 或省略且命中未过期、满足该来源缓存新鲜度的 Dataset 时直接返回已有 DatasetRef，不访问上游；force_refresh=true 才重新取数并生成新的不可变 Dataset（默认保留 7 天），绝不覆盖旧 Dataset。明确要现在/最新的行情快照时请用 force_refresh=true，7 天是保留期而非实时新鲜度。本工具只返回 DatasetRef（dataset_id / artifact_ref / schema / row_count / captured_at / retention_until 等元数据），绝不返回原始数据行。capability 必须是 list_capabilities 返回的短能力名（如 ticker_search / quote / history / trading_calendar），params 必须按 describe_capability({ capability }) 返回的 input_schema 填写。若返回 request_params_invalid，先重新调用 describe_capability({ capability: "<当前能力名>" })，再按 schema 修正并重试一次；不要凭记忆猜参数。task_id 必须由主 Agent 在本轮委派前确定，并在所有子 Agent 与最终报告中原样复用；不要由 data_collector 自行生成。相同 capability+params 的并发请求合并为同一次执行。当前 workspace 只读或落盘失败时抛错 workspace_not_writable（绝不退化为内存假成功）。请求归属自动记为当前 delegated child session，不接收 requester_agent_id 参数。', jsonObject({
+            capability: { type: 'string' },
+            params: freeObject,
+            force_refresh: { type: 'boolean' },
+            task_id: { type: 'string' },
+        }, ['capability', 'params', 'task_id']), datasetRefSchema(), async (args, exec) => {
+            const requestedCapability = boundedString(args.capability, 'capability', MAX_ID_LENGTH);
+            const session = delegatedSession(exec, 'request_data');
+            try {
+                const params = boundedParams(args.params);
+                // Fail early with the same catalog guidance as describe_capability instead of
+                // making the model infer a capability typo from a low-level Hub error.
+                const capability = resolveCapability(hub, requestedCapability);
+                return await hub.request({
+                    capability,
+                    params,
+                    force_refresh: args.force_refresh === true,
+                    task_id: boundedString(args.task_id, 'task_id', MAX_ID_LENGTH),
+                    session,
+                }, { signal: exec.signal });
+            }
+            catch (error) {
+                throw requestParamsError(error, requestedCapability) ?? error;
+            }
+        }),
+        tool('list_capabilities', '列出当前已注册的数据能力目录：**纯文本，一行一条** `capability|summary|paginated`，paginated 为 1（分页）/ 0（不分页）。名字或摘要里的 `|`、换行会转义成 `\\|`、`\\n`，所以一行就是一条能力，不要按冒号或空格拆分。目录是发现入口，**一个任务只需调用一次**：同一 session 的历史里已经保有这份目录，重复调用只会白占上下文。目录刻意不含参数与输出结构（完整 schema 约 23KB，会被剪枝截断）；确定要用的能力后，用 describe_capability 单独取那一个能力的 input_schema 后即可 request_data。request_data 的 capability 必须以本目录为准，不要自行编造。', jsonObject(), { type: 'string' }, async (_args, exec) => { delegatedSession(exec, 'list_capabilities'); return hub.capabilityDirectory(); }),
+        tool('describe_capability', '读取**一个**能力的完整契约：description（单位、null 语义、时间与分页口径）、input_schema（params 的取值与约束）、output_fields（返回字段的逐行字典 `路径:类型[:说明]`，类型后缀 `?` 为可为 null、`[].` 为数组元素的字段）与 paginated。request_data 的 params 必须按这里返回的 input_schema 填写；request_data 若返回 request_params_invalid，应回到这里重新读取后再重试。一个能力描述一次即可：结果已在本 session 历史里，重复描述同一个能力只会白占上下文；需要几个能力就分别调用几次，不要用逗号把多个名字塞进一次调用。错误按 code 引导：capability_required（没给名字，先 list_capabilities()）、capability_invalid（不是字符串/超长/含逗号，改为目录中的单个名字）、capability_catalog_empty（数据源未注册，别重试，回告主 Agent）、capability_unknown（名字不存在，错误里会列出全部可用能力名并要求原样复制）。', jsonObject({ capability: { type: 'string' } }, ['capability']), jsonObject({
+            capability: { type: 'string' },
+            summary: { type: 'string' },
+            paginated: { type: 'boolean' },
+            description: { type: 'string' },
+            input_schema: freeObject,
+            output_fields: { oneOf: [{ type: 'string' }, { type: 'null' }] },
+        }, ['capability', 'summary', 'paginated', 'description', 'input_schema', 'output_fields']), async (args, exec) => {
+            delegatedSession(exec, 'describe_capability');
+            const capability = resolveCapability(hub, args.capability);
+            const detail = hub.describeCapability(capability);
+            if (!detail)
+                throw capabilityError('capability_unknown', `未注册的能力 "${capability}"。`);
+            return detail;
+        }),
+        ...(diagnostics ? [
+            tool('dc_status', '诊断工具：返回数据管道运行状态 —— at/call 为本次真实执行的时间戳与计数（防伪）、api_key 为**本次调用现取**的凭据解析结果（present/source，不含密钥值）、registered_capabilities 为当前已注册 capability 列表、root_tool_policy 为根 Agent 工具收敛的现场记录（preset_id / outcome / denied / failed，排查"主 Agent 为什么看得见本该收敛掉的工具"用）。数据源注册不依赖 key：api_key.present=false 时同花顺能力仍在目录里，但每次取数会点名 FUYAO_API_KEY 失败。排障时优先调用。', jsonObject(), jsonObject({
+                at: { type: 'integer' },
+                call: { type: 'integer' },
+                api_key: jsonObject({ present: { type: 'boolean' }, source: { oneOf: [{ type: 'string' }, { type: 'null' }] }, error: { oneOf: [{ type: 'string' }, { type: 'null' }] } }, ['present']),
+                registered_capabilities: { type: 'array', items: { type: 'string' } },
+                root_tool_policy: { type: 'array', items: jsonObject({
+                        agent_id: { oneOf: [{ type: 'string' }, { type: 'null' }] },
+                        root: { type: 'boolean' },
+                        preset_id: { type: 'string' },
+                        outcome: { type: 'string' },
+                        denied: { type: 'array', items: { type: 'string' } },
+                        failed: { type: 'array', items: { type: 'string' } },
+                    }, ['agent_id', 'root', 'preset_id', 'outcome', 'denied', 'failed']) },
+            }, ['at', 'call', 'api_key', 'registered_capabilities']), async (_args, exec) => {
+                delegatedSession(exec, 'dc_status');
+                const apiKey = await diagnostics.probeApiKey();
+                const capabilities = hub.capabilityNames();
+                const rootPolicy = diagnostics.rootPolicy?.()?.map((probe) => ({
+                    agent_id: probe.agentId ?? null,
+                    root: probe.root,
+                    preset_id: probe.presetId,
+                    outcome: probe.outcome,
+                    denied: probe.denied,
+                    failed: probe.failed,
+                }));
+                const snapshot = { at: Date.now(), call: ++dcStatusCalls.current, apiKey, capabilities };
+                diagnostics.onCall?.(snapshot);
+                return {
+                    at: snapshot.at,
+                    call: snapshot.call,
+                    api_key: apiKey,
+                    registered_capabilities: capabilities,
+                    ...(rootPolicy === undefined ? {} : { root_tool_policy: rootPolicy }),
+                };
+            }),
+        ] : []),
+    ];
+    for (const definition of registrations)
+        ctx.effect(() => tools.register(definition), `capital-generation.tool(${definition.name})`);
+}
+/** 供测试与宿主侧断言复用：模型可见的 DatasetRef 字段白名单。 */
+export const DATASET_REF_FIELDS = [
+    'dataset_id', 'task_id', 'session_id', 'artifact_ref', 'format',
+    'capability', 'source_label', 'schema', 'row_count', 'captured_at',
+    'retention_until', 'params_digest',
+];

package/lib/index.d.ts

@@ -0,0 +1,275 @@
+import { Context } from '@deepseek-ai/cordis';
+import z from '@deepseek-ai/schemastery';
+/** Internal plugin name used by the Capital mode preset. */
+export declare const name = "capital-generation";
+/**
+ * 取数配置所在的 **profile 条目 id**（随包的 `capital-config` 行）。0.1.7 的 settings 面
+ * 以条目 id 为命名空间，本插件经 `settings.describe()` 按它读回用户配的引用名与开关。
+ *
+ * 这个字符串有三处必须一致：`cordis.patch.yml` 的行 id、`capital-config/index.js` 的
+ * `SETTINGS_ENTRY_ID`、以及浏览器半边 `client.src.cjs` 的 `ENTRY_ID`；由
+ * `test/capital-config.test.mjs` 逐一对齐（漂移的表现为卡片静默消失）。
+ */
+export declare const CAPITAL_CONFIG_ENTRY_ID = "capital-config";
+/** Wind 文档检索配置：JSON-RPC 端点与凭据名（线格式为 MCP 线协议，零依赖适配）。 */
+export interface WindDocsConfig {
+    endpoint?: string;
+    credentialRef?: string;
+    timeoutMs?: number;
+}
+/** 本地直连回退配置（AnySearch 明确失败后改由本机直连抓取公开文本页面）。 */
+export interface LocalFetchConfig {
+    enabled?: boolean;
+    timeoutMs?: number;
+    maxBytes?: number;
+    maxContentChars?: number;
+    maxRedirects?: number;
+    userAgent?: string;
+}
+/**
+ * PaddleOCR 文档解析（`ocr` 工具）配置。
+ *
+ * `endpoint` / `model` 空值 = 用 `src/ocr/client.ts` 里的官方端点与模型名（真值只在那一处，
+ * 与 `windDocs.endpoint` 同一口径）。`pollBudgetMs` 是**工具内部**等作业的自有预算，
+ * 必须小于工具声明的 `timeoutMs: 240000`——余量留给"决定回 pending 之后把回执写出去"，
+ * 撞上框架超时就没有 `doc_id` 可续查了。
+ */
+export interface PaddleOcrConfig {
+    endpoint?: string;
+    credentialRef?: string;
+    model?: string;
+    pollBudgetMs?: number;
+}
+/** web_retriever 会话配置：anysearch（广度）+ wind_docs（public_document 精准）+ 文档解析。 */
+export interface RetrieverConfig {
+    baseURL?: string;
+    credentialRef?: string;
+    windDocs?: WindDocsConfig;
+    localFetch?: LocalFetchConfig;
+    paddleOcr?: PaddleOcrConfig;
+}
+/**
+ * 本地直连回退的默认值（唯一真值来源）：schema 的 `.default()` 与消费点
+ * `resolveLocalFetchConfig()` 都引用它，两处不再各写一份字面量。
+ *
+ * 数值口径对齐官方 `dsh-web-fetch-http`（DSH 自己的同类本机抓取器）：
+ * - `timeoutMs` = 30000（官方默认）。2026-09 实测：环境代理的故障转移慢路径约
+ *   15.2–15.4s，15s 预算会把证券业协会官网这类站点误判成 TIMEOUT。
+ * - `maxContentChars` = 100000（官方 `maxBodyChars` 默认值）。注意它切的是**转换前的
+ *   原始 HTML**（见 `html-markdown.ts`），而 markdown 输出只有 HTML 的 12–33%，
+ *   所以这个数必须按 HTML 体积给足；20,000 会把 109KB 的页面腰斩到 18%。
+ *
+ * `userAgent` 默认是产品标识（`@v587d/capital-generation`）：裸版本号是
+ * WAF 眼里的典型爬虫特征。显式配成空串时消费点回落到 `LOCAL_FETCH_CLIENT_VERSION`
+ * （版本号真值仍只有一处，由测试守着等于 `package.json` 的 version）。
+ */
+export declare const LOCAL_FETCH_DEFAULTS: Required<LocalFetchConfig>;
+/**
+ * `ocr`（PaddleOCR 文档解析）配置的默认值：与 `LOCAL_FETCH_DEFAULTS` 同一口径，
+ * schema 的 `.default()` 与消费点都引用它，字面量只有一处。
+ *
+ * `endpoint` / `model` 留空是把真值让给 `src/ocr/client.ts`（那边还有 `PADDLE_OCR_MAX_PAGES`
+ * 这类跟着端点走的常数）；`pollBudgetMs: 0` 同理让给客户端的 `DEFAULT_WAIT_BUDGET_MS`。
+ */
+export declare const PADDLE_OCR_DEFAULTS: Required<PaddleOcrConfig>;
+/**
+ * 消费点独立补齐默认值。`apply()` 在无 settings 的宿主/测试里拿到的是**未过 schema**
+ * 的原始对象，`retriever.localFetch` 可能是 `undefined`，因此消费点不能依赖 schema 补默认值。
+ */
+export declare function resolveLocalFetchConfig(config?: LocalFetchConfig): Required<LocalFetchConfig>;
+/** Configuration accepted by the Capital Generation plugin. */
+export interface Config {
+    /** Optional additive persona override; core safety guidance is preserved. */
+    customPersona?: string;
+    /** Fuyao credentials 引用名；空值回退到 FUYAO_API_KEY。 */
+    fuyaoCredentialRef?: string;
+    /** web_retriever 配置（可选；缺省使用 AnySearch 默认地址与凭据名）。 */
+    retriever?: RetrieverConfig;
+}
+/** DSH 0.1.2-rc.1 configuration schema. */
+export declare const Config: z<Schemastery.ObjectS<NoInfer<{
+    customPersona: z<string, string, "defined">;
+    fuyaoCredentialRef: z<string, string, "defined">;
+    retriever: z<Schemastery.ObjectS<NoInfer<{
+        baseURL: z<string, string, "defined">;
+        credentialRef: z<string, string, "defined">;
+        windDocs: z<Schemastery.ObjectS<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            timeoutMs: z<number, number, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            timeoutMs: z<number, number, "defined">;
+        }>>, "defined">;
+        localFetch: z<Schemastery.ObjectS<NoInfer<{
+            enabled: z<boolean, boolean, "defined">;
+            timeoutMs: z<number, number, "defined">;
+            maxBytes: z<number, number, "defined">;
+            maxContentChars: z<number, number, "defined">;
+            maxRedirects: z<number, number, "defined">;
+            userAgent: z<string, string, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            enabled: z<boolean, boolean, "defined">;
+            timeoutMs: z<number, number, "defined">;
+            maxBytes: z<number, number, "defined">;
+            maxContentChars: z<number, number, "defined">;
+            maxRedirects: z<number, number, "defined">;
+            userAgent: z<string, string, "defined">;
+        }>>, "defined">;
+        paddleOcr: z<Schemastery.ObjectS<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            model: z<string, string, "defined">;
+            pollBudgetMs: z<number, number, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            model: z<string, string, "defined">;
+            pollBudgetMs: z<number, number, "defined">;
+        }>>, "defined">;
+    }>>, Schemastery.ObjectT<NoInfer<{
+        baseURL: z<string, string, "defined">;
+        credentialRef: z<string, string, "defined">;
+        windDocs: z<Schemastery.ObjectS<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            timeoutMs: z<number, number, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            timeoutMs: z<number, number, "defined">;
+        }>>, "defined">;
+        localFetch: z<Schemastery.ObjectS<NoInfer<{
+            enabled: z<boolean, boolean, "defined">;
+            timeoutMs: z<number, number, "defined">;
+            maxBytes: z<number, number, "defined">;
+            maxContentChars: z<number, number, "defined">;
+            maxRedirects: z<number, number, "defined">;
+            userAgent: z<string, string, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            enabled: z<boolean, boolean, "defined">;
+            timeoutMs: z<number, number, "defined">;
+            maxBytes: z<number, number, "defined">;
+            maxContentChars: z<number, number, "defined">;
+            maxRedirects: z<number, number, "defined">;
+            userAgent: z<string, string, "defined">;
+        }>>, "defined">;
+        paddleOcr: z<Schemastery.ObjectS<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            model: z<string, string, "defined">;
+            pollBudgetMs: z<number, number, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            model: z<string, string, "defined">;
+            pollBudgetMs: z<number, number, "defined">;
+        }>>, "defined">;
+    }>>, "defined">;
+}>>, Schemastery.ObjectT<NoInfer<{
+    customPersona: z<string, string, "defined">;
+    fuyaoCredentialRef: z<string, string, "defined">;
+    retriever: z<Schemastery.ObjectS<NoInfer<{
+        baseURL: z<string, string, "defined">;
+        credentialRef: z<string, string, "defined">;
+        windDocs: z<Schemastery.ObjectS<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            timeoutMs: z<number, number, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            timeoutMs: z<number, number, "defined">;
+        }>>, "defined">;
+        localFetch: z<Schemastery.ObjectS<NoInfer<{
+            enabled: z<boolean, boolean, "defined">;
+            timeoutMs: z<number, number, "defined">;
+            maxBytes: z<number, number, "defined">;
+            maxContentChars: z<number, number, "defined">;
+            maxRedirects: z<number, number, "defined">;
+            userAgent: z<string, string, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            enabled: z<boolean, boolean, "defined">;
+            timeoutMs: z<number, number, "defined">;
+            maxBytes: z<number, number, "defined">;
+            maxContentChars: z<number, number, "defined">;
+            maxRedirects: z<number, number, "defined">;
+            userAgent: z<string, string, "defined">;
+        }>>, "defined">;
+        paddleOcr: z<Schemastery.ObjectS<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            model: z<string, string, "defined">;
+            pollBudgetMs: z<number, number, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            model: z<string, string, "defined">;
+            pollBudgetMs: z<number, number, "defined">;
+        }>>, "defined">;
+    }>>, Schemastery.ObjectT<NoInfer<{
+        baseURL: z<string, string, "defined">;
+        credentialRef: z<string, string, "defined">;
+        windDocs: z<Schemastery.ObjectS<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            timeoutMs: z<number, number, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            timeoutMs: z<number, number, "defined">;
+        }>>, "defined">;
+        localFetch: z<Schemastery.ObjectS<NoInfer<{
+            enabled: z<boolean, boolean, "defined">;
+            timeoutMs: z<number, number, "defined">;
+            maxBytes: z<number, number, "defined">;
+            maxContentChars: z<number, number, "defined">;
+            maxRedirects: z<number, number, "defined">;
+            userAgent: z<string, string, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            enabled: z<boolean, boolean, "defined">;
+            timeoutMs: z<number, number, "defined">;
+            maxBytes: z<number, number, "defined">;
+            maxContentChars: z<number, number, "defined">;
+            maxRedirects: z<number, number, "defined">;
+            userAgent: z<string, string, "defined">;
+        }>>, "defined">;
+        paddleOcr: z<Schemastery.ObjectS<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            model: z<string, string, "defined">;
+            pollBudgetMs: z<number, number, "defined">;
+        }>>, Schemastery.ObjectT<NoInfer<{
+            endpoint: z<string, string, "defined">;
+            credentialRef: z<string, string, "defined">;
+            model: z<string, string, "defined">;
+            pollBudgetMs: z<number, number, "defined">;
+        }>>, "defined">;
+    }>>, "defined">;
+}>>, "plain">;
+/**
+ * 把可选的用户人设文本转成独立 system-prompt section（官方 systemPrompt
+ * registry 的注册对象）。空白输入返回 undefined（不注册）；超长抛错；
+ * 追加不可覆盖的安全提醒。只影响表达风格与呈现。
+ */
+export declare function resolveUserCustomizationSection(customPersona?: string): {
+    name: string;
+    order: number;
+    text: string;
+} | undefined;
+/** The system-prompt service is a hard dependency for the optional section. */
+export declare const inject: string[];
+/**
+ * Register the Capital mode data services and tools.
+ *
+ * 人设文本全部由 preset 组合承载（dsh-persona 行 + 两个委派行 config.persona），
+ * 不在插件代码中。本包只装配服务与工具：
+ *  - datasetStore（workspace-local Dataset 持久化，7 天保留，权限失败显式报错）
+ *  - dataCollectorHub（阻塞 FIFO + in-flight 合并，成功后由 store 立即落盘，
+ *    只回传 DatasetRef，不保留长期 raw 缓存）
+ *  - 无状态 AnySearch 网页检索（单查询 search + 单 URL fetch）
+ *  - 无状态 Wind 文档检索客户端（public_document：公告/年报/招股书与权威新闻）
+ *  - 各自的模型工具；时间工具。
+ */
+export declare function apply(ctx: Context, config: Config): void;