feat: add plugin manager and official plugins (#119)

* feat: add plugin manager and official plugins * fix(agent-core): honor plugin capability overrides * fix: restrict plugin zip root detection * Update apps/kimi-code/src/constant/app.ts Co-authored-by: liruifengv <liruifeng1024@gmail.com> Signed-off-by: qer <wbxl2000@outlook.com> --------- Signed-off-by: qer <wbxl2000@outlook.com> Co-authored-by: liruifengv <liruifeng1024@gmail.com>
2026-07-09 17:29:12 +00:00 · 2026-05-27 22:47:33 +08:00 · 2026-05-27 22:47:33 +08:00 · ebf6e8181e
commit ebf6e8181e
parent 2c7a8cc010
75 changed files with 7327 additions and 23 deletions
--- a/plugins/marketplace.json
+++ b/plugins/marketplace.json
@ -0,0 +1,24 @@
+{
+  "version": "1",
+  "plugins": [
+    {
+      "id": "kimi-datasource",
+      "tier": "official",
+      "displayName": "Kimi Datasource",
+      "version": "3.0.0",
+      "description": "Official datasource workflows.",
+      "keywords": ["data", "mcp"],
+      "source": "./official/kimi-datasource"
+    },
+    {
+      "id": "superpowers",
+      "tier": "curated",
+      "displayName": "Superpowers",
+      "version": "5.1.0",
+      "description": "Planning, TDD, debugging, and delivery workflows for coding agents.",
+      "homepage": "https://github.com/obra/superpowers",
+      "keywords": ["skills", "planning", "tdd", "debugging", "code-review"],
+      "source": "./curated/superpowers"
+    }
+  ]
+}
--- a/plugins/official/kimi-datasource/.gitignore
+++ b/plugins/official/kimi-datasource/.gitignore
@ -0,0 +1,6 @@
+# 编辑器
+.vscode/
+.idea/
+*.swp
+*.swo
+*~
--- a/plugins/official/kimi-datasource/SKILL.md
+++ b/plugins/official/kimi-datasource/SKILL.md
@ -0,0 +1,150 @@
+---
+name: kimi-datasource
+description: |
+  通用数据源助手。当用户要查股票/财报/技术指标/全球宏观经济/中国企业工商/学术论文这类外部数据时，使用这个 skill。
+  本 plugin 通过 MCP server `plugin-kimi-datasource-data` 提供工具；优先调用 `mcp__plugin-kimi-datasource-data__query_stock`、`mcp__plugin-kimi-datasource-data__get_data_source_desc`、`mcp__plugin-kimi-datasource-data__call_data_source_tool`。
+---
+
+# kimi-datasource — 通用数据源助手
+
+## 0. 调用方式
+
+本 skill 使用 datasource MCP server 注册的三个工具，不要通过 Bash 手动执行脚本：
+
+- `mcp__plugin-kimi-datasource-data__query_stock`
+- `mcp__plugin-kimi-datasource-data__get_data_source_desc`
+- `mcp__plugin-kimi-datasource-data__call_data_source_tool`
+
+这三个工具由 Kimi Code 托管执行，参数直接按 tool schema 传 JSON。
+
+工具会读取 `$KIMI_CODE_HOME/credentials/kimi-code.json`。如果没有登录凭据，让用户先在 Kimi Code 里执行 `/login`。
+
+## 1. 这个 skill 提供什么能力
+
+本 plugin 后面挂了 6 个外部数据源。每一行的"数据源名"就是传给 `get_data_source_desc` 的 `name`。
+
+| 能力域 | 数据源名 | 典型问题 |
+|---|---|---|
+| **A股 / 港股 / 美股 行情和财务** | `stock_finance_data` | "茅台现在多少钱"、"宁德时代 2024 年财报"、"腾讯股东"、"杭州的人工智能股票" |
+| **Yahoo Finance 全球金融** | `yahoo_finance` | "苹果分析师评级"、"AAPL 期权链"、"标普 500 历年价格" |
+| **世界银行宏观经济** | `world_bank_open_data` | "中国历年 GDP"、"印度通胀率"、"各国人口增长对比" |
+| **中国企业工商信息** | `tianyancha` | "字节跳动股东"、"比亚迪司法风险"、"宁德时代专利" |
+| **arXiv 论文预印本** | `arxiv` | "找 RAG 综述"、"下载 2406.xxxxx" |
+| **Google Scholar 学术搜索** | `scholar` | "Hinton 最新论文"、"transformer 综述高引文献" |
+
+**不支持的能力**：通用 Web 搜索 / 实时新闻。问到这类问题，告诉用户当前数据源不覆盖。
+
+## 2. 标准工作流：`get_data_source_desc` → `call_data_source_tool`
+
+后端可用 API 经常会调整，**这份 skill 故意不抄具体的 API 名和参数表**。每次调用前你都应当现场问数据源："你都有什么接口？"
+
+```
+1. 根据用户问题，从上表挑出一个 data_source_name
+2. 执行 get_data_source_desc，读取该数据源的 Markdown 文档
+3. 仔细读返回的 Markdown，里面列了：
+     - 该数据源整体说明（含 ticker 格式、全局约束）
+     - 每个 API 的描述 / 必填参数 / 可选参数 / 默认值 / 取值范围
+4. 选最匹配的 API，按文档拼 params
+5. 执行 call_data_source_tool
+6. 读返回结果给用户
+```
+
+### 例 1：用户问"茅台最近一年走势"
+
+1. 股票走势 → `stock_finance_data`
+2. 调用 `mcp__plugin-kimi-datasource-data__get_data_source_desc`，参数 `{"name":"stock_finance_data"}`
+
+3. 从文档里找到"获取历史价格"那个 API，看它要 `ticker / start_date / end_date / file_path` 等
+4. 用 web_search 核对 → 茅台 = `600519.SH`
+5. 调用 `mcp__plugin-kimi-datasource-data__call_data_source_tool`，参数形如 `{"data_source_name":"stock_finance_data","api_name":"<文档里写的 api>","params":{"ticker":"600519.SH","start_date":"...","end_date":"...","file_path":"/tmp/mao_1y.csv"}}`
+
+### 例 2：用户问"找几篇 retrieval augmented generation 的综述"
+
+1. 论文搜索 → `arxiv`（或 `scholar`，arxiv 更适合预印本，scholar 引用更全）
+2. 调用 `mcp__plugin-kimi-datasource-data__get_data_source_desc`，参数 `{"name":"arxiv"}`
+
+3. 从文档里找到搜索类 API，看它要 `query / file_path / max_results` 等
+4. 执行 `call_data_source_tool`
+
+### 例 3：用户问"字节跳动有哪些股东"
+
+1. 企业工商 → `tianyancha`
+2. 调用 `mcp__plugin-kimi-datasource-data__get_data_source_desc`，参数 `{"name":"tianyancha"}`
+
+3. 注意：tianyancha 的 API 是动态注册的，文档会指引你**先用搜索类接口找到合适的 API 名，再调用**
+4. **必须使用企业全称**（"北京字节跳动科技有限公司"），不要用简称。不知道全称就先用 tianyancha 文档里的"公司搜索"接口查
+
+## 3. 调用前的几条铁律
+
+### 3.1 股票代码必须核对，不能凭记忆猜
+
+A 股 `.SH/.SZ/.BJ`，港股 `.HK`，美股 `.US` 等。用户通常只说中文名（"茅台"、"宁德时代"、"腾讯"），不会给代码。
+
+**调任何股票相关 API 前**，先用 `web_search` / `WebSearch` 一类联网工具确认正确代码 + 后缀。
+
+如果当前环境没有任何联网工具，**让用户亲口确认代码**，不要硬猜。错了的话接口会静默返回错数据或空数据。
+
+### 3.2 企业相关查询必须用全称
+
+`tianyancha` 拒收"特斯拉"、"网易"、"腾讯"这种简称，必须给"北京特斯拉销售有限公司"这种全名。不知道全名时，先调它的公司搜索 API。
+
+### 3.3 多数 API 需要 `file_path`
+
+绝大部分数据源 API 把完整结果以 CSV 形式写到 `file_path`。漏传会报 `Missing required parameters: file_path`。不知道传啥时，给一个 `/tmp/<场景>_<时间戳>.csv` 即可。
+
+### 3.4 一次调用不要堆太多 ticker
+
+`stock_finance_data` 的实时接口最多 3 个 ticker，历史接口最多 10 个。超过会被截断或报错。多了就分批调。
+
+## 4. 怎么读返回结果
+
+`call_data_source_tool` 和 `query_stock` 的 stdout 一般含两段：
+
+1. **`data_preview`**：CSV 头 + 前几行（通常 1~3 行），方便你直接答简单问题
+2. **`CSV 数据已写入：/tmp/xxx.csv`**：完整数据落盘路径
+
+策略：
+- 用户只问"XX 现在多少钱"、"中国 2023 GDP 多少"这种单值 → `data_preview` 一般够，直接答
+- 用户要画图、对比、算盈亏、列清单 → 用 `Read` 工具把 CSV 读出来再处理
+- 混合 A+港股查询时服务端会自动把 CSV 拆成 `_a.csv` / `_hk.csv` 两份，原 `file_path` 那个文件不存在
+
+如果接口返回失败，提示文字一般会写明原因（参数不对 / 不支持 / 数据空等）。把人话原因反馈给用户，不要硬走第二次。
+
+## 5. `query_stock` — 实时行情快捷命令
+
+对**实时**类的常见股票查询，可以**不走** `get_data_source_desc → call_data_source_tool` 这套流程，直接用 `query_stock`，省一次调用。它等价于 `stock_finance_data` 的实时接口子集。
+
+调用 `mcp__plugin-kimi-datasource-data__query_stock`，参数形如 `{"ticker":"600519.SH","type":"realtime_price","file_path":"/tmp/stock_600519.csv"}`。
+
+适用场景（且仅限）：
+- 看当前价 / 当日分钟 K 线（`type=realtime_price`）
+- 看实时技术指标 MA/MACD/KDJ/RSI/BOLL 等（`type=realtime_tech`，**仅 A 股**，港股/美股会报错）
+- 开盘摘要（`type=open_summary`）
+- 收盘摘要（`type=close_summary`，**港股盘后看当天数据必须用这个**，`realtime_price` 拿不到）
+
+涉及**历史日 K / 财报 / 公告 / 股东 / 财务指标 / 选股 / 业务分部 / 预测**等任何"非实时"的股票查询，走标准 `get_data_source_desc → call_data_source_tool` 流程，不要在 `query_stock` 上硬凑。
+
+## 6. `watchlist.json` — 用户自选股
+
+`${KIMI_SKILL_DIR}/watchlist.json` 是用户的自选股列表。用户问"看一下我的自选股"时，读这个文件然后按每 3 只一组分批执行 `query_stock`。
+
+格式：
+
+```json
+[
+  {"code": "600519.SH", "name": "贵州茅台"},
+  {"code": "0700.HK", "name": "腾讯控股", "hold_cost": 350.5, "hold_quantity": 100}
+]
+```
+
+- `code` 和 `name` 必填；`hold_cost` 和 `hold_quantity` 可选
+- 两者都有时顺便算盈亏：`(当前价 - hold_cost) * hold_quantity`
+- 用户说"帮我加 XX 到自选股"时：先 web_search 核对代码，再追加到 JSON 数组
+
+## 7. 注意事项
+
+- **不要凭记忆猜股票代码 / 企业全称**。错代码会让接口静默返回错数据，用户察觉不到
+- **不要在没读 desc 的情况下硬传 `api_name`**。后端会报 `API_NOT_FOUND`。除非这次会话里你已经读过该数据源的 desc 并记得参数
+- **不要给投资建议**。给完数据加一句"AI 生成，不构成投资建议"即可
+- **不要在 `query_stock` 上塞历史/财报查询**。它只覆盖实时类，别的会失败
+- 如果某个数据源接口返回的报错明显是后端 bug（参数 schema 自相矛盾、内部 Python 报错等），**汇报错误给用户，不要硬试**——这种 bug 我们这边修不了，要后端服务侧改
--- a/plugins/official/kimi-datasource/bin/kimi-datasource.mjs
+++ b/plugins/official/kimi-datasource/bin/kimi-datasource.mjs
@ -0,0 +1,446 @@
+#!/usr/bin/env node
+// Stdio MCP server for kimi-datasource.
+//
+// Speaks newline-delimited JSON-RPC 2.0 on stdin/stdout per the MCP "stdio"
+// transport. Implements the minimal surface the Kimi Code host calls:
+//   - initialize
+//   - notifications/initialized
+//   - tools/list
+//   - tools/call
+//   - ping
+//
+// Business logic (API call, credentials, headers) is unchanged from the
+// previous one-shot CLI; only the transport changed.
+
+import { randomUUID } from 'node:crypto';
+import { mkdir, readFile, writeFile } from 'node:fs/promises';
+import { arch, homedir, hostname, release, type } from 'node:os';
+import path from 'node:path';
+import readline from 'node:readline';
+
+const VERSION = '3.0.0';
+const API_URL = process.env.KIMI_DATASOURCE_API_URL ?? 'https://api.kimi.com/coding/v1/tools';
+const REQUEST_TIMEOUT_MS = 30_000;
+const PROTOCOL_VERSION = '2025-06-18';
+const VALID_STOCK_QUERY_TYPES = new Set([
+  'realtime_price',
+  'realtime_tech',
+  'open_summary',
+  'close_summary',
+]);
+
+const TOOLS = [
+  {
+    name: 'query_stock',
+    description:
+      'Query realtime stock price, realtime technical indicators, open summaries, or close summaries for up to 3 tickers.',
+    inputSchema: {
+      type: 'object',
+      properties: {
+        ticker: {
+          type: 'string',
+          description: 'Ticker code list separated by commas, for example 600519.SH or 0700.HK.',
+        },
+        type: {
+          type: 'string',
+          enum: ['realtime_price', 'realtime_tech', 'open_summary', 'close_summary'],
+          description: 'Realtime stock query type.',
+        },
+        time: {
+          type: 'string',
+          description: 'Optional time parameter for supported realtime endpoints.',
+        },
+        file_path: {
+          type: 'string',
+          description: 'Optional CSV output path. When omitted, the tool chooses a temporary path.',
+        },
+      },
+      required: ['ticker'],
+    },
+  },
+  {
+    name: 'get_data_source_desc',
+    description:
+      'Get the current API documentation for one Kimi data source before calling a specific API.',
+    inputSchema: {
+      type: 'object',
+      properties: {
+        name: {
+          type: 'string',
+          enum: [
+            'stock_finance_data',
+            'yahoo_finance',
+            'world_bank_open_data',
+            'tianyancha',
+            'arxiv',
+            'scholar',
+          ],
+          description: 'Data source name.',
+        },
+      },
+      required: ['name'],
+    },
+  },
+  {
+    name: 'call_data_source_tool',
+    description: 'Call one API from a Kimi data source after reading get_data_source_desc.',
+    inputSchema: {
+      type: 'object',
+      properties: {
+        data_source_name: {
+          type: 'string',
+          description: 'Data source name returned or documented by get_data_source_desc.',
+        },
+        api_name: {
+          type: 'string',
+          description: 'API name from the data source description.',
+        },
+        params: {
+          type: 'object',
+          description: 'API parameters that match the data source description.',
+        },
+      },
+      required: ['data_source_name', 'api_name', 'params'],
+    },
+  },
+];
+
+const HANDLERS = {
+  query_stock: {
+    method: 'get_stock_realtime_price',
+    buildParams(args) {
+      const ticker = requiredString(args, 'ticker');
+      const tickerList = ticker
+        .split(',')
+        .map((item) => item.trim())
+        .filter(Boolean);
+      if (tickerList.length === 0) throw new Error('Missing required argument: ticker.');
+      if (tickerList.length > 3) {
+        throw new Error('ticker accepts at most 3 values separated by commas.');
+      }
+
+      const queryType = optionalString(args, 'type') ?? 'realtime_price';
+      if (!VALID_STOCK_QUERY_TYPES.has(queryType)) {
+        throw new Error(
+          `type must be one of ${JSON.stringify([...VALID_STOCK_QUERY_TYPES])}; received: ${queryType}`,
+        );
+      }
+
+      const params = {
+        ticker,
+        type: queryType,
+        file_path: optionalString(args, 'file_path') ?? defaultStockFilePath(ticker, queryType),
+      };
+      const time = optionalString(args, 'time');
+      if (time !== undefined) params.time = time;
+      return params;
+    },
+    format(text, params) {
+      return `${text}\n\nCSV data written to: ${params.file_path}`;
+    },
+  },
+  get_data_source_desc: {
+    method: 'get_data_source_desc',
+    buildParams(args) {
+      return { name: requiredString(args, 'name') };
+    },
+  },
+  call_data_source_tool: {
+    method: 'call_data_source_tool',
+    buildParams(args) {
+      return {
+        data_source_name: requiredString(args, 'data_source_name'),
+        api_name: requiredString(args, 'api_name'),
+        params: requiredObject(args, 'params'),
+      };
+    },
+  },
+};
+
+async function handleRequest(message) {
+  const { method, id, params } = message;
+  switch (method) {
+    case 'initialize':
+      return {
+        protocolVersion: PROTOCOL_VERSION,
+        capabilities: { tools: {} },
+        serverInfo: { name: 'kimi-datasource', version: VERSION },
+      };
+    case 'ping':
+      return {};
+    case 'tools/list':
+      return { tools: TOOLS };
+    case 'tools/call':
+      return runTool(params);
+    default:
+      throw jsonRpcError(-32601, `Method not found: ${method}`, { id });
+  }
+}
+
+async function runTool(params) {
+  const name = params?.name;
+  const args = params?.arguments ?? {};
+  const handler = HANDLERS[name];
+  if (handler === undefined) {
+    return {
+      content: [{ type: 'text', text: `Unknown tool: ${String(name)}` }],
+      isError: true,
+    };
+  }
+  try {
+    const built = handler.buildParams(args);
+    const response = await callKimiTool(handler.method, built);
+    const text = extractText(response);
+    const formatted = (handler.format?.(text, built) ?? text).trim();
+    return { content: [{ type: 'text', text: formatted }] };
+  } catch (err) {
+    const message = err instanceof Error ? err.message : String(err);
+    return {
+      content: [{ type: 'text', text: message }],
+      isError: true,
+    };
+  }
+}
+
+function resolveKimiHome() {
+  const explicit = process.env.KIMI_CODE_HOME?.trim();
+  return explicit && explicit.length > 0 ? explicit : path.join(homedir(), '.kimi-code');
+}
+
+async function loadAccessToken() {
+  const kimiHome = resolveKimiHome();
+  const credentialsFile = path.join(kimiHome, 'credentials', 'kimi-code.json');
+  let parsed;
+  try {
+    parsed = JSON.parse(await readFile(credentialsFile, 'utf8'));
+  } catch (err) {
+    if (isNotFound(err)) {
+      throw new Error(
+        `Kimi Code credentials file not found: ${credentialsFile}\nRun /login in Kimi Code first.`,
+      );
+    }
+    if (err instanceof SyntaxError) {
+      throw new Error(`Failed to parse Kimi Code credentials file: ${err.message}`);
+    }
+    throw err;
+  }
+
+  if (!isRecord(parsed)) {
+    throw new Error(`Invalid Kimi Code credentials file: ${credentialsFile}`);
+  }
+  const token = typeof parsed.access_token === 'string' ? parsed.access_token : '';
+  if (token.length === 0) {
+    throw new Error('Kimi Code credentials do not contain access_token. Run /login again.');
+  }
+  const expiresAt = typeof parsed.expires_at === 'number' ? parsed.expires_at : 0;
+  if (expiresAt > 0 && expiresAt <= Math.floor(Date.now() / 1000)) {
+    throw new Error('Kimi Code access_token has expired. Run /login again and retry.');
+  }
+  return { kimiHome, token };
+}
+
+async function callKimiTool(method, params) {
+  const { kimiHome, token } = await loadAccessToken();
+  const controller = new AbortController();
+  const timeout = setTimeout(() => {
+    controller.abort();
+  }, REQUEST_TIMEOUT_MS);
+  try {
+    const response = await fetch(API_URL, {
+      method: 'POST',
+      headers: await buildHeaders(kimiHome, token),
+      body: JSON.stringify({ method, params }),
+      signal: controller.signal,
+    });
+    const text = await response.text();
+    if (!response.ok) {
+      throw new Error(`HTTP ${response.status} error: ${text}`);
+    }
+    try {
+      return JSON.parse(text);
+    } catch {
+      return text;
+    }
+  } catch (err) {
+    if (err instanceof DOMException && err.name === 'AbortError') {
+      throw new Error(`Request timed out after ${REQUEST_TIMEOUT_MS / 1000} seconds.`);
+    }
+    throw err;
+  } finally {
+    clearTimeout(timeout);
+  }
+}
+
+async function buildHeaders(kimiHome, token) {
+  return {
+    Authorization: `Bearer ${token}`,
+    'Content-Type': 'application/json',
+    'X-Msh-Tool-Call-Id': randomUUID(),
+    'X-Msh-Platform': asciiHeader(process.env.KIMI_MSH_PLATFORM ?? 'kimi-code-cli'),
+    'X-Msh-Version': asciiHeader(process.env.KIMI_MSH_VERSION ?? VERSION),
+    'X-Msh-Device-Name': asciiHeader(process.env.KIMI_MSH_DEVICE_NAME ?? hostname()),
+    'X-Msh-Device-Model': asciiHeader(process.env.KIMI_MSH_DEVICE_MODEL ?? deviceModel()),
+    'X-Msh-Os-Version': asciiHeader(process.env.KIMI_MSH_OS_VERSION ?? release()),
+    'X-Msh-Device-Id': asciiHeader(process.env.KIMI_MSH_DEVICE_ID ?? (await createDeviceId(kimiHome))),
+    'User-Agent': `kimi-datasource/${VERSION}`,
+  };
+}
+
+async function createDeviceId(kimiHome) {
+  const deviceIdPath = path.join(kimiHome, 'device_id');
+  try {
+    const existing = (await readFile(deviceIdPath, 'utf8')).trim();
+    if (existing.length > 0) return existing;
+  } catch {
+    // Fall through to create a best-effort local device id.
+  }
+
+  const id = randomUUID();
+  try {
+    await mkdir(kimiHome, { recursive: true, mode: 0o700 });
+    await writeFile(deviceIdPath, `${id}\n`, { encoding: 'utf8', mode: 0o600 });
+  } catch {
+    // Headers can still use the in-memory id if the file cannot be written.
+  }
+  return id;
+}
+
+function deviceModel() {
+  const os = type();
+  const osVersion = release();
+  const osArch = arch();
+  if (os === 'Darwin') return `macOS ${osVersion} ${osArch}`;
+  if (os === 'Windows_NT') return `Windows ${osVersion} ${osArch}`;
+  return `${os} ${osVersion} ${osArch}`.trim();
+}
+
+function extractText(response) {
+  if (typeof response === 'string') return response;
+  if (!isRecord(response)) return String(response);
+
+  if (response.is_success === false) {
+    const message = extractUserText(response.error) ?? JSON.stringify(response);
+    throw new Error(`Tool API returned an error: ${message}`);
+  }
+
+  const text = extractUserText(response.result);
+  if (text !== undefined) return text;
+  return `Tool API succeeded but did not return user text. Raw response: ${JSON.stringify(response)}`;
+}
+
+function extractUserText(value) {
+  if (!isRecord(value) || !Array.isArray(value.user)) return undefined;
+  const text = value.user
+    .filter((item) => isRecord(item) && item.type === 'text' && typeof item.text === 'string')
+    .map((item) => item.text)
+    .filter(Boolean)
+    .join('\n\n');
+  return text.length > 0 ? text : undefined;
+}
+
+function defaultStockFilePath(ticker, queryType) {
+  const safeTicker = ticker.replaceAll(',', '_').replaceAll('.', '_');
+  return `/tmp/stock_${safeTicker}_${queryType}.csv`;
+}
+
+function requiredString(args, field) {
+  const value = optionalString(args, field);
+  if (value === undefined) throw new Error(`Missing required argument: ${field}.`);
+  return value;
+}
+
+function optionalString(args, field) {
+  if (!isRecord(args)) return undefined;
+  const value = args[field];
+  if (value === undefined || value === null) return undefined;
+  if (typeof value !== 'string') throw new Error(`${field} must be a string.`);
+  const trimmed = value.trim();
+  return trimmed.length > 0 ? trimmed : undefined;
+}
+
+function requiredObject(args, field) {
+  if (!isRecord(args)) throw new Error(`Missing required argument: ${field}.`);
+  const value = args[field];
+  if (!isRecord(value)) throw new Error(`${field} must be an object.`);
+  return value;
+}
+
+function isRecord(value) {
+  return typeof value === 'object' && value !== null && !Array.isArray(value);
+}
+
+function isNotFound(err) {
+  return isRecord(err) && err.code === 'ENOENT';
+}
+
+function asciiHeader(value, fallback = 'unknown') {
+  const cleaned = String(value).replaceAll(/[^ -~]/g, '').trim();
+  return cleaned.length > 0 ? cleaned : fallback;
+}
+
+function jsonRpcError(code, message, data) {
+  const err = new Error(message);
+  err.jsonRpc = { code, message, data };
+  return err;
+}
+
+function send(message) {
+  process.stdout.write(`${JSON.stringify(message)}\n`);
+}
+
+function sendResult(id, result) {
+  send({ jsonrpc: '2.0', id, result });
+}
+
+function sendError(id, error) {
+  send({ jsonrpc: '2.0', id, error });
+}
+
+async function dispatch(message) {
+  if (message?.jsonrpc !== '2.0') return;
+  // Notifications carry no id and never expect a response.
+  if (message.id === undefined || message.id === null) {
+    if (message.method === 'notifications/initialized' || message.method === 'notifications/cancelled') {
+      return;
+    }
+    return;
+  }
+  const id = message.id;
+  try {
+    const result = await handleRequest(message);
+    sendResult(id, result ?? {});
+  } catch (err) {
+    if (err && typeof err === 'object' && err.jsonRpc !== undefined) {
+      sendError(id, err.jsonRpc);
+      return;
+    }
+    sendError(id, {
+      code: -32603,
+      message: err instanceof Error ? err.message : String(err),
+    });
+  }
+}
+
+function start() {
+  const rl = readline.createInterface({ input: process.stdin });
+  rl.on('line', (line) => {
+    const trimmed = line.trim();
+    if (trimmed.length === 0) return;
+    let message;
+    try {
+      message = JSON.parse(trimmed);
+    } catch (err) {
+      sendError(null, {
+        code: -32700,
+        message: `Parse error: ${err instanceof Error ? err.message : String(err)}`,
+      });
+      return;
+    }
+    void dispatch(message);
+  });
+  rl.on('close', () => {
+    process.exit(0);
+  });
+}
+
+start();
--- a/plugins/official/kimi-datasource/kimi.plugin.json
+++ b/plugins/official/kimi-datasource/kimi.plugin.json
@ -0,0 +1,18 @@
+{
+  "name": "kimi-datasource",
+  "version": "3.0.0",
+  "description": "Finance, macro, enterprise, and academic data tools for Kimi Code.",
+  "keywords": ["finance", "data-source", "mcp"],
+  "mcpServers": {
+    "data": {
+      "command": "node",
+      "args": ["./bin/kimi-datasource.mjs"],
+      "cwd": "./"
+    }
+  },
+  "interface": {
+    "displayName": "Kimi Datasource",
+    "shortDescription": "Finance, macro, enterprise, and academic data tools",
+    "developerName": "Moonshot AI"
+  }
+}
--- a/plugins/official/kimi-datasource/watchlist.json
+++ b/plugins/official/kimi-datasource/watchlist.json
@ -0,0 +1,14 @@
+[
+  {
+    "code": "600519.SH",
+    "name": "贵州茅台"
+  },
+  {
+    "code": "000001.SZ",
+    "name": "平安银行"
+  },
+  {
+    "code": "0700.HK",
+    "name": "腾讯控股"
+  }
+]