agent-bridge-pa 0.0.0-stage → 1.0.7

package/README.md

@@ -1,3 +1,68 @@
-# Temporary Holding Version
+# 个人助理管理 Agent
 
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.
+`agent-bridge-pa` 是通过 AUN 使用的管理 Agent，运行时插件 ID 仍为 `personal-assistant`。安装并启动插件后，先创建一个专用 AID，再把该 AID 绑定到 `personal-assistant`；普通聊天和管理操作都在这个 AUN 会话中完成。
+
+## 配置
+
+`endpoint`（完整的 OpenAI 兼容 `/chat/completions` 地址）、`model` 和 `apiKey` 是自然语言工具调度的可选配置。未配置模型时，精确的 `/assistant agent call <agent-id> <任务>` 命令仍可使用核心授权直接转发。`allowedPeers`、`allowedAgents` 默认均为 `[]`（界面留空），接受所有 AUN Peer，允许调用五个桌面 Agent、CLI 及后续安装的普通 Agent。绑定个人助理 AID 后无需进入 ABWeb 设置白名单即可测试，目标 Agent 仍须已就绪；自身和其他管理型 Agent 不可作为派活目标。两项名单留空接受所有，填写具体身份/插件 ID 后才限制；旧 `*` 配置保留兼容。`timeoutMs` 控制单次模型请求，默认 120 秒。密钥只保存在本机配置中，不要写入仓库或消息。白名单留空的规则要求核心 `>=1.0.78`，独立插件会拒绝不满足版本的旧核心。
+
+需要限制时，在 ABWeb 的个人助理配置中将 `allowedPeers` 改为指定的 AID、将 `allowedAgents` 改为允许的插件 ID。清空名单即恢复全部允许，保存和重载后保持此规则。升级保留已有具体名单；缺失字段或空数组均表示全部允许，无需填写 `*`。模型来源及 AID 绑定继续沿用原配置。
+
+个人助理要求支持插件默认配置的核心 `mu-agent-bridge >=1.0.74`，插件使用 Node.js 22.16 或更高版本。仅有旧核心版本时会拒绝加载不兼容的包。
+
+在 ABWeb 的“已安装插件 → 个人助理 → 配置模型”选择来源。插件列表来自 daemon 的能力接口，包含已声明模型来源的五个桌面 Agent（WorkBuddy、千问办公、豆包、DSH、ChatGPT）及 Codex CLI、Claude CLI；未就绪的插件保留显示并给出原因，不会自动替换已选来源。选择“OpenAI 兼容接口”时填写地址和模型名；本地免密服务可以留空密钥。保存表单只修改有变化的字段，空白密钥保留原值，隐藏的 Owner、能力及 AID 配置不会被覆盖。
+
+插件来源配置示例：
+
+```json
+{
+  "modelProvider": "plugin",
+  "modelPlugin": "qwenwork",
+  "timeoutMs": 120000
+}
+```
+
+上述示例省略白名单，使用默认的所有 AUN Peer 和所有已就绪的普通 Agent。
+
+`modelPlugin` 决定个人助理使用哪个插件生成决策，`allowedAgents` 决定允许向哪些 Agent 派发任务，二者独立。桌面来源沿用桌面已登录账号和当前模型，无需填写 API Key；CLI 沿用 daemon 用户的默认模型。WorkBuddy 桌面来源需要 `transport: "cdp"`；CLI 需要先配置独立的绝对工作目录并登录。来源插件及核心都需更新到支持通用模型入口的版本。
+
+同一 AUN 会话的模型上下文保存在个人助理数据目录的 `model-sessions.json` 中，没有按闲置时长失效的计时器。重载后会继续读取；在普通私聊发送 `/new` 会生成新的会话，不带入旧内容（话题内 `/new` 沿用核心的拒绝规则）。最多保留最近 128 个会话，每会话 40 条消息及 120000 个序列化字符，超限按完整轮次删除旧历史。历史包含用户正文和工具结果，请按聊天记录保护本地文件；不写入模型配置密钥或附件二进制。写盘失败会记录警告并保留内存历史，但此时重载可能丢失尚未落盘的内容。
+
+模型会话使用原通道、接收 AID、Peer 与 thread 的逻辑会话及来源 ID 隔离；桌面映射保存在原插件数据目录中，重载后可恢复。模型请求与普通任务共用原插件队列，不发送聊天心跳，不为调用模型启动或重启桌面；用户有草稿、正在生成或等待确认时会拒绝切换。取消先结束个人助理等待；桌面仍在生成时继续占用队列，确认生成结束后才释放，不把迟到正文送入后续会话。
+
+模型返回的 JSON 继续经过个人助理原六轮、八次工具调用上限及参数、工具名、白名单校验，不把 Bridge 管理对象、通道回调或附件交给模型来源。失败回复区分来源、阶段和错误码，固定宿主原因可见，任意异常正文不直接回显。
+
+桌面模型来源沿用客户端已有的文件和工具权限，Bridge 无法统一关闭这些原生能力。个人助理的管理白名单只约束 Bridge 管理工具，不代表桌面客户端运行在无工具环境中。
+
+也可设置 `modelProvider: "claude-cli"`，复用本机 `claude-cli` 插件的默认模型和登录，无需填写 `endpoint`、`model`、`apiKey`；未设置来源时仍使用原 OpenAI 兼容接口。先为 Claude CLI 插件配置已存在的独立绝对工作目录 `cwd`，并在 daemon 的运行用户下完成 CLI 登录。个人助理通过插件内部接口请求结构化决策，不传 `--model`，由 CLI 的用户配置决定默认模型。
+
+Claude 模型请求禁用 CLI 原生工具、MCP 和 hooks，且不保存 CLI 会话；管理工具仍由个人助理按既有授权与白名单执行。普通 Claude CLI 聊天保留原行为。Claude 模式还要求核心提供 `modelInputGuard` 隔离能力保证；旧核心会使个人助理降级并提示升级，不会调用 CLI。CLI 需要支持 `--json-schema` 等隔离参数，不支持时会明确失败，不能回退为普通 CLI 任务。
+
+远程能力与协调还需配置 `selfAid`（个人助理 AID）和 `ecPath`（可选，默认 `ec`）。`abilityProviders` 明确列出可访问的 Provider；首期只允许声明为纯只读且无需认证/确认的 Ability。pending 请求只能查询，不会自动重发。
+
+跨 Agent 派活沿用 `allowedAgents`，并要求目标 Agent 已就绪且 AID 与核心报告的绑定一致。每次派活会创建独立 thread；发送回执只表示已受理，需用 `coordination_status` 查询 thread history。进程只保留最近 128 个本机派活 thread 的查询关联，不会自动重派。
+
+Owner 共享资料通过 `ownerAid` 与唯一的规范绝对路径 `ownerContextPath` 配置，缓存有效期由 `ownerContextTtlMs` 控制（默认 60 秒，最多 5 分钟）。资料会作为参考内容注入模型，不具备指令或授权效力。可直接发送 `/assistant context refresh` 强制重读；读取失败会保留之前成功的缓存。未设置 Owner 配置时不读取共享资料。
+
+## 创建和绑定专用 AID
+
+先用原有 CLI 或管理后台创建并绑定专用身份；个人助理尚未绑定时无法接收自己的管理命令。创建时必须提供 Owner AID：
+
+```text
+mab aun create assistant.agentid.cn --owner owner.agentid.cn
+mab aun bind assistant.agentid.cn personal-assistant
+```
+
+创建失败不会自动重试，也不会在回复中显示密钥库路径、seed 或 token。绑定成功后，将 AUN 配置中的 `agent_plugin` 指向 `personal-assistant`，若配置了非空 `allowedPeers`，需确保发消息的 Peer 在名单中。个人助理投入使用后，可用 `/assistant aid create <aid> [--owner <owner-aid>]` 和 `/assistant aid bind <aid> <agent-id> [--replace]` 显式管理后续身份；省略 Owner 时沿用 AUN 的本地默认 Owner。`--replace` 只替换目标 Agent 自己原有关联；已经绑定到其他 Agent 的 AID 一律拒绝转移，请先通过原管理命令解除旧绑定。
+
+## 能力
+
+经授权的请求可以查询 Agent 能力和 AID、调用白名单中的 Agent、查询 AUN 身份、执行只读 `/sessions` 命令、生成今天工作的小结，以及查看个人助理状态。模型只能通过固定工具访问这些能力，不能提交 URL、Peer、session 或 principal 覆盖字段。
+
+需要确定性地转发给某个已授权 Agent 时，可直接发送 `/assistant agent call <agent-id> <任务>`，例如 `/assistant agent call workbuddy 请仅回复联调标记`。核心会再次校验 Agent 白名单和就绪状态；此命令不经过模型工具判断。
+
+用自然语言询问今天的工作小结时，个人助理通过 `sessions_today` 读取已归档记录，归档采集默认关闭。未归档记录和原生插件历史不会纳入该小结；显式 `/sessions today --summary` 按核心命令读取插件提供的原生历史。会话查询的 `allowFrom` 默认 `[]`（留空接受所有），各项查询和小结均允许读取本机其他 Peer 的记录，旧配置缺少此字段也按全部允许处理。需要限制时设置非空的具体 Peer 名单，名单外的来源使用精确的 `grants`/`nativeGrants`。清单编号仍绑定当前来源和逻辑会话；`/new` 后重新列举即可，默认无需再次授权。只读总结器使用核心独立的 summarizer 配置，不会调用普通 Agent 执行总结。
+
+个人助理仅支持文本输入。模型端点必须是本地管理员配置的 HTTPS 或 HTTP 地址，插件禁止跟随重定向；模型响应和工具结果都有大小及调用次数限制。
+
+子 Agent 返回的真实附件随个人助理最终回复回传，模型仅看到附件元数据；是否发送成功由通道确认。即使后续模型调用失败，已经取得的附件也会随错误说明返回。

package/config.schema.json

@@ -0,0 +1,23 @@
+{
+  "type": "object",
+  "default": { "allowedPeers": [], "allowedAgents": [] },
+  "properties": {
+    "modelProvider": { "type": "string", "enum": ["openai", "claude-cli", "plugin"], "default": "openai", "description": "模型来源；plugin 使用已安装插件的当前模型" },
+    "modelPlugin": { "type": "string", "pattern": "^[a-z][a-z0-9-]{1,63}$", "description": "模型来源插件 ID" },
+    "endpoint": { "type": "string", "minLength": 1, "maxLength": 2048, "pattern": "^https?://[^\\s]+/chat/completions$", "description": "OpenAI 兼容 chat/completions 完整地址" },
+    "aid": { "type": "string", "maxLength": 256, "pattern": "^(?:[A-Za-z0-9][A-Za-z0-9._-]+)?$", "description": "绑定到个人助理的 AUN AID，由核心在绑定时维护；空字符串表示解绑" },
+    "model": { "type": "string", "minLength": 1, "maxLength": 256, "description": "模型名称" },
+    "apiKey": { "type": "string", "minLength": 1, "maxLength": 4096, "secret": true, "writeOnly": true, "description": "模型服务密钥，仅保存在本机配置中" },
+    "allowedPeers": { "type": "array", "uniqueItems": true, "items": { "type": "string", "pattern": "^(?:\\*|[A-Za-z0-9][A-Za-z0-9._-]{2,255})$" }, "default": [], "description": "留空接受所有 AUN Peer；填写具体 AID 后限制来源" },
+    "allowedAgents": { "type": "array", "uniqueItems": true, "items": { "type": "string", "pattern": "^(?:\\*|[a-z0-9][a-z0-9-]{0,63})$" }, "default": [], "description": "留空允许调用所有已安装且就绪的普通 Agent；填写具体插件 ID 后限制目标" },
+    "timeoutMs": { "type": "integer", "minimum": 1000, "maximum": 120000, "default": 120000, "description": "单次模型请求最长时间（毫秒）" },
+    "ecPath": { "type": "string", "minLength": 1, "maxLength": 2048, "description": "EvolCore ec 命令路径，默认使用 PATH 中的 ec" },
+    "selfAid": { "type": "string", "maxLength": 256, "pattern": "^(?:[A-Za-z0-9][A-Za-z0-9._-]+)?$", "description": "个人助理自身的 AUN AID；未配置时关闭 Ability 和协调能力" },
+    "abilityProviders": { "type": "array", "uniqueItems": true, "items": { "type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._-]{2,255}$" }, "default": [], "description": "允许读取与调用只读 Ability 的 Provider AID 白名单" },
+    "ownerAid": { "type": "string", "maxLength": 256, "pattern": "^(?:[A-Za-z0-9][A-Za-z0-9._-]+)?$", "description": "Owner 共享上下文的固定来源 AID" },
+    "ownerContextPath": { "type": "string", "maxLength": 2048, "pattern": "^/[^\\\\]*$", "description": "Owner AUN FS 中固定的规范绝对路径" },
+    "ownerContextTtlMs": { "type": "integer", "minimum": 1000, "maximum": 300000, "default": 60000, "description": "Owner 共享上下文缓存有效期（毫秒）" }
+  },
+  "required": ["allowedPeers", "allowedAgents"],
+  "additionalProperties": false
+}

package/dist/src-ts/agent.js

@@ -0,0 +1,69 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.AgentOperationError = exports.AgentError = void 0;
+exports.abortable = abortable;
+exports.errorDetails = errorDetails;
+async function abortable(promise, signal) {
+    if (!signal)
+        return promise;
+    if (signal.aborted) {
+        // promise 已在调用方启动，先观察其拒绝，避免预取消时产生未处理拒绝。
+        void promise.catch(() => undefined);
+        signal.throwIfAborted();
+    }
+    return new Promise((resolve, reject) => {
+        let settled = false;
+        const cleanup = () => signal.removeEventListener("abort", abort);
+        const abort = () => {
+            if (settled)
+                return;
+            settled = true;
+            cleanup();
+            reject(signal.reason);
+        };
+        signal.addEventListener("abort", abort, { once: true });
+        promise.then((value) => {
+            if (settled)
+                return;
+            settled = true;
+            cleanup();
+            resolve(value);
+        }, (error) => {
+            if (settled)
+                return;
+            settled = true;
+            cleanup();
+            reject(error);
+        });
+    });
+}
+class AgentError extends Error {
+    constructor(message, options) {
+        super(message, options);
+        this.name = "AgentError";
+    }
+}
+exports.AgentError = AgentError;
+/** 保留底层错误码，避免只留下 fetch failed 或超时这类表层提示。 */
+function errorDetails(error) {
+    const parts = [];
+    const seen = new Set();
+    for (let current = error; current && !seen.has(current) && seen.size < 5;) {
+        seen.add(current);
+        const value = current;
+        const message = typeof value.message === "string" ? value.message : String(current);
+        parts.push(`${message}${value.code ? `（${String(value.code)}）` : ""}`);
+        current = value.cause;
+    }
+    return parts.join("；原因：") || String(error);
+}
+class AgentOperationError extends AgentError {
+    code;
+    stage;
+    constructor(code, stage, cause) {
+        super(`${code}：${stage}失败：${errorDetails(cause)}`, { cause });
+        this.code = code;
+        this.stage = stage;
+    }
+}
+exports.AgentOperationError = AgentOperationError;

package/dist/src-ts/attachments.js

@@ -0,0 +1,62 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.MAX_TOTAL_ATTACHMENT_BYTES = exports.MAX_ATTACHMENTS = exports.MAX_ATTACHMENT_BYTES = void 0;
+exports.detectImageContentType = detectImageContentType;
+exports.attachmentFilename = attachmentFilename;
+exports.validateAttachment = validateAttachment;
+exports.validateAttachments = validateAttachments;
+exports.asAttachments = asAttachments;
+exports.MAX_ATTACHMENT_BYTES = 32 * 1024 * 1024;
+exports.MAX_ATTACHMENTS = 8;
+exports.MAX_TOTAL_ATTACHMENT_BYTES = 64 * 1024 * 1024;
+function detectImageContentType(data) {
+    if (data.length >= 8 && data.slice(0, 8).every((value, index) => value === [137, 80, 78, 71, 13, 10, 26, 10][index]))
+        return "image/png";
+    if (data.length >= 3 && data[0] === 0xff && data[1] === 0xd8 && data[2] === 0xff)
+        return "image/jpeg";
+    if (data.length >= 12 && String.fromCharCode(...data.slice(0, 4)) === "RIFF" && String.fromCharCode(...data.slice(8, 12)) === "WEBP")
+        return "image/webp";
+    return undefined;
+}
+function attachmentFilename(filename) {
+    if (typeof filename !== "string" || /[\u0000-\u001f\u007f]/.test(filename))
+        throw new Error("附件文件名无效");
+    const name = filename.split(/[\\/]/).pop() ?? "";
+    if (!name.trim() || name === "." || name === "..")
+        throw new Error("附件文件名无效");
+    return name;
+}
+function validateAttachment(item) {
+    if (!item || !(item.data instanceof Uint8Array) || item.data.byteLength === 0)
+        throw new Error("附件内容不能为空");
+    if (item.data.byteLength > exports.MAX_ATTACHMENT_BYTES)
+        throw new Error("单个附件不能超过 32 MiB");
+    attachmentFilename(item.filename);
+    if (item.kind !== "image" && item.kind !== "file")
+        throw new Error("附件类型无效");
+    if (typeof item.contentType !== "string" || !/^[A-Za-z0-9!#$&^_.+-]+\/[A-Za-z0-9!#$&^_.+-]+$/.test(item.contentType))
+        throw new Error("附件 MIME 类型无效");
+    if (item.kind === "image") {
+        const detected = detectImageContentType(item.data);
+        if (!detected || detected !== item.contentType.toLowerCase())
+            throw new Error("图片内容与 MIME 类型不一致");
+    }
+}
+function validateAttachments(items) {
+    if (!items)
+        return;
+    if (!Array.isArray(items))
+        throw new Error("附件列表无效");
+    if (items.length > exports.MAX_ATTACHMENTS)
+        throw new Error("单条消息最多包含 8 个附件");
+    let total = 0;
+    for (const item of items) {
+        validateAttachment(item);
+        total += item.data.byteLength;
+    }
+    if (total > exports.MAX_TOTAL_ATTACHMENT_BYTES)
+        throw new Error("单条消息附件总大小不能超过 64 MiB");
+}
+function asAttachments(images) {
+    return (images ?? []).map((image) => ({ ...image, kind: "image" }));
+}

package/dist/src-ts/plugin.js

@@ -0,0 +1,13 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.resolveRuntimePort = resolveRuntimePort;
+function resolveRuntimePort(config, runtime, pluginId, envName, fallback) {
+    const injected = runtime?.ports?.[pluginId];
+    if (typeof injected === "number" && Number.isInteger(injected) && injected > 0)
+        return injected;
+    const configured = config.cdpPort;
+    if (typeof configured === "number" && Number.isInteger(configured) && configured > 0)
+        return configured;
+    const environment = Number(process.env[envName] ?? "");
+    return Number.isInteger(environment) && environment > 0 ? environment : fallback;
+}

package/dist/src-ts/plugins/personal-assistant/claude-cli-model.js

@@ -0,0 +1,30 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.responseSchema = void 0;
+exports.claudeCliCompletion = claudeCliCompletion;
+const tools_1 = require("./tools");
+exports.responseSchema = {
+    type: "object", additionalProperties: false, required: ["role", "content", "tool_calls"],
+    properties: {
+        role: { type: "string", enum: ["assistant"] },
+        content: { type: ["string", "null"], maxLength: 32_000 },
+        tool_calls: { type: "array", maxItems: 8, items: {
+                type: "object", additionalProperties: false, required: ["id", "type", "function"],
+                properties: {
+                    id: { type: "string", minLength: 1, maxLength: 128 }, type: { type: "string", enum: ["function"] },
+                    function: { type: "object", additionalProperties: false, required: ["name", "arguments"], properties: {
+                            name: { type: "string", enum: [...tools_1.TOOL_NAMES] }, arguments: { type: "string", maxLength: 64_000 },
+                        } },
+                },
+            } },
+    },
+};
+function claudeCliCompletion(callPlugin, timeoutMs) {
+    return async (messages, signal) => {
+        const response = await callPlugin("claude-cli", {
+            task: JSON.stringify({ instructions: "你是个人助理的模型推理器。按照 messages 中的系统要求处理最后一条用户消息；tools 仅描述宿主提供的管理工具。需要查询或执行任务时返回 tool_calls，待工具结果返回后再回答；不要编造结果。每次调用使用不同 id。最终回答时 tool_calls 为空数组。仅返回 Schema 规定的 JSON。", messages, tools: (0, tools_1.toolDefinitions)() }),
+            modelResponseSchema: exports.responseSchema, timeoutMs, signal,
+        });
+        return response.text;
+    };
+}

package/dist/src-ts/plugins/personal-assistant/coordination.js

@@ -0,0 +1,57 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.createCoordinator = createCoordinator;
+const node_child_process_1 = require("node:child_process");
+const node_util_1 = require("node:util");
+const execFileAsync = (0, node_util_1.promisify)(node_child_process_1.execFile);
+const isAid = (value) => typeof value === "string" && /^[A-Za-z0-9][A-Za-z0-9._-]{2,255}$/.test(value);
+const isAgentId = (value) => typeof value === "string" && /^[a-z0-9][a-z0-9-]{0,63}$/.test(value);
+const isThreadId = (value) => typeof value === "string" && /^[A-Za-z0-9][A-Za-z0-9._:-]{0,255}$/.test(value);
+function createCoordinator(options) {
+    if (!isAid(options.selfAid))
+        throw new Error("selfAid 无效");
+    const allowed = new Map();
+    for (const agent of options.allowedAgents) {
+        if (!isAgentId(agent.id) || !Array.isArray(agent.aids) || agent.aids.some((aid) => !isAid(aid)))
+            throw new Error("Agent 白名单无效");
+        allowed.set(agent.id, new Set(agent.aids));
+    }
+    const run = options.execFile ?? (async (file, args) => execFileAsync(file, args, { timeout: 30_000, maxBuffer: 1024 * 1024, encoding: "utf8", windowsHide: true }));
+    const command = async (args) => {
+        const { stdout } = await run(options.ecPath ?? "ec", args);
+        let value;
+        try {
+            value = JSON.parse(stdout);
+        }
+        catch {
+            throw new Error("EvolCore 返回了无效 JSON");
+        }
+        if (!value || typeof value !== "object" || Array.isArray(value) || value.ok === false)
+            throw new Error("EvolCore 命令失败");
+        return value;
+    };
+    return {
+        async dispatch({ agentId, aid, task }) {
+            if (!isAgentId(agentId) || !isAid(aid) || !allowed.get(agentId)?.has(aid))
+                throw new Error("目标 Agent/AID 不在白名单中");
+            if (typeof task !== "string" || !task.trim() || task.length > 20_000)
+                throw new Error("任务长度无效");
+            const thread = await command(["thread", "create", aid, "--name", `个人助理派活-${agentId}`, "--format", "json"]);
+            const threadId = thread.thread_id ?? thread.threadId ?? thread.id;
+            if (!isThreadId(threadId))
+                throw new Error("EvolCore 未返回有效 threadId");
+            await command(["msg", "send", options.selfAid, aid, "--text", task, "--thread", threadId, "--return", "required", "--format", "json"]);
+            return { status: "pending", agentId, aid, threadId };
+        },
+        async getStatus({ aid, threadId }) {
+            if (!isAid(aid))
+                throw new Error("AID 无效");
+            if (!isThreadId(threadId))
+                throw new Error("threadId 无效");
+            const history = await command(["thread", "history", threadId, "--format", "json"]);
+            const state = history.status ?? history.state;
+            const status = state === "completed" || state === "failed" || state === "cancelled" ? state : "pending";
+            return { status, aid, threadId, history };
+        },
+    };
+}

package/dist/src-ts/plugins/personal-assistant/index.js

@@ -0,0 +1,263 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.PersonalAssistantPlugin = exports.personalAssistantManifest = void 0;
+const attachments_1 = require("../../attachments");
+const agent_1 = require("../../agent");
+const node_crypto_1 = require("node:crypto");
+const model_client_1 = require("./model-client");
+const tools_1 = require("./tools");
+const remote_ability_1 = require("./remote-ability");
+const owner_context_1 = require("./owner-context");
+const claude_cli_model_1 = require("./claude-cli-model");
+const session_memory_1 = require("./session-memory");
+exports.personalAssistantManifest = {
+    id: "personal-assistant",
+    version: "1.0.7",
+    type: "agent",
+    displayName: "个人助理",
+    entry: "dist/src-ts/plugins/personal-assistant/index.js",
+    apiVersion: "1",
+    process: "in-process",
+    capabilities: ["chat", "management", "health"],
+    configSchema: "config.schema.json",
+};
+class PersonalAssistantPlugin {
+    manifest = exports.personalAssistantManifest;
+    sessionOperations = ["new"];
+    context;
+    client;
+    state = "failed";
+    stateMessage = "插件尚未启动";
+    fetcher;
+    ownerContext;
+    threads = new Map();
+    constructor(options = {}) { this.fetcher = options.fetch; }
+    async activate(context) {
+        this.context = context;
+        const config = context.config;
+        try {
+            const endpoint = validEndpoint(config.endpoint);
+            if (!Array.isArray(config.allowedPeers) || !Array.isArray(config.allowedAgents) || !config.allowedPeers.every(value => value === "*" || isAid(value)) || !config.allowedAgents.every(value => value === "*" || isPluginId(value)))
+                throw new Error("allowedPeers 或 allowedAgents 配置无效");
+            if (config.modelProvider !== undefined && !["openai", "claude-cli", "plugin"].includes(config.modelProvider))
+                throw new Error("不支持该模型来源");
+            const claudeMode = config.modelProvider === "claude-cli";
+            const pluginMode = config.modelProvider === "plugin";
+            const modelPlugin = pluginMode && isPluginId(config.modelPlugin) ? config.modelPlugin : undefined;
+            if (claudeMode && context.modelInputGuard !== true)
+                throw new Error("当前核心不支持安全的 Claude 模型调用，请升级核心");
+            if (pluginMode && !modelPlugin)
+                throw new Error("请选择有效的模型来源插件");
+            if (pluginMode && typeof context.callModelProvider !== "function")
+                throw new Error("当前核心不支持通用模型来源，请升级核心");
+            const modelConfigured = claudeMode || pluginMode || Boolean(config.endpoint || config.model || config.apiKey);
+            if (!claudeMode && !pluginMode && modelConfigured && (!endpoint || !config.model?.trim()))
+                throw new Error("请配置有效的模型 endpoint 和 model");
+            const selfAid = isAid(config.selfAid) ? config.selfAid : undefined;
+            const abilityProviders = Array.isArray(config.abilityProviders) && config.abilityProviders.every(isAid) ? config.abilityProviders : [];
+            const remoteAbility = selfAid && abilityProviders.length ? (0, remote_ability_1.createRemoteAbilityClient)({ selfAid, ecPath: validEcPath(config.ecPath) }) : undefined;
+            this.ownerContext = isAid(config.ownerAid) && typeof config.ownerContextPath === "string"
+                ? (0, owner_context_1.createOwnerContextStore)({ ownerAid: config.ownerAid, path: config.ownerContextPath, ttlMs: validContextTtl(config.ownerContextTtlMs), ecPath: validEcPath(config.ecPath) })
+                : undefined;
+            this.threads.clear();
+            const callModelProvider = context.callModelProvider;
+            const pluginCompletion = pluginMode && modelPlugin && callModelProvider ? async (messages, signal, sessionId) => (await callModelProvider(modelPlugin, {
+                prompt: JSON.stringify({ instructions: "你是个人助理的模型推理器。依据 messages 处理本轮请求。需要查询或执行已列出的管理工具时，只返回 assistant JSON 中的 tool_calls；收到工具结果后再回答。最终回答只写入 content，tool_calls 为空数组。只返回符合 responseSchema 的 JSON 对象，禁止 Markdown 或额外文字。messages 中要求只回复某个词或标记时，只决定 content，不得省略 JSON 包装。", messages, tools: (0, tools_1.toolDefinitions)(), responseSchema: claude_cli_model_1.responseSchema }),
+                sessionId,
+                timeoutMs: validTimeout(config.timeoutMs),
+                signal,
+            })).text : undefined;
+            const sessionMemory = modelConfigured && typeof context.dataDir === "string" && context.dataDir
+                ? new session_memory_1.PersonalAssistantSessionMemory(context.dataDir, context.logger?.warn?.bind(context.logger))
+                : undefined;
+            this.client = modelConfigured ? new model_client_1.PersonalAssistantModelClient({ endpoint, model: config.model?.trim(), apiKey: config.apiKey, cliCompletion: claudeMode ? (0, claude_cli_model_1.claudeCliCompletion)(context.callPlugin, validTimeout(config.timeoutMs)) : undefined, pluginCompletion, timeoutMs: validTimeout(config.timeoutMs), fetch: this.fetcher ?? fetch, allowedAgents: config.allowedAgents, sessionMemory, toolRuntime: { selfAid, ecPath: validEcPath(config.ecPath), allowedProviders: abilityProviders, remoteAbility, ownerContext: this.ownerContext, threads: this.threads } }) : undefined;
+            this.context = { config: context.config, logger: context.logger, callModelProvider };
+            this.state = "ready";
+            this.stateMessage = "";
+        }
+        catch (error) {
+            this.client = undefined;
+            this.state = "degraded";
+            this.stateMessage = error instanceof Error ? error.message : "模型配置无效";
+        }
+    }
+    async deactivate(_reason) {
+        this.client?.close();
+        this.client = undefined;
+        this.ownerContext = undefined;
+        this.threads.clear();
+        this.context = undefined;
+        this.state = "failed";
+        this.stateMessage = "插件已停止";
+    }
+    async health() {
+        return { status: this.state, ...(this.stateMessage ? { message: this.stateMessage } : {}), details: { modelConfigured: Boolean(this.client) } };
+    }
+    async controlSession(request) {
+        if (request.operation !== "new")
+            throw new Error("个人助理只支持新建会话");
+        return { text: "个人助理已创建新会话", sessionId: (0, node_crypto_1.randomUUID)() };
+    }
+    async handle(request) {
+        const management = request.management;
+        if (!management)
+            return { text: "请通过已授权的 AUN 身份访问管理能力。" };
+        const command = (0, tools_1.parseExplicitCommand)(request.task);
+        if (command?.kind === "invalid")
+            return { text: "不支持该管理命令，请检查命令格式后重试。" };
+        if (command?.kind === "sessions") {
+            try {
+                return { text: await management.runSessionCommand(command.command) };
+            }
+            catch {
+                return { text: "会话管理命令执行失败，请检查权限和命令格式。" };
+            }
+        }
+        if (command?.kind === "context-refresh") {
+            if (!this.ownerContext)
+                return { text: "Owner 共享上下文尚未配置。" };
+            try {
+                const refreshed = await this.ownerContext.refresh();
+                return { text: `Owner 共享上下文已刷新（${refreshed.content.length} 字符）。` };
+            }
+            catch {
+                return { text: "Owner 共享上下文刷新失败，已保留之前成功读取的缓存。" };
+            }
+        }
+        if (command?.kind === "agent-call") {
+            try {
+                const result = await management.callAgent(command.agentId, command.task);
+                return { ...result, sessionId: request.sessionId };
+            }
+            catch {
+                return { text: `调用 Agent ${command.agentId} 失败，请检查白名单和目标状态。` };
+            }
+        }
+        if (command?.kind === "create") {
+            try {
+                return { text: safeJson(await management.createAunIdentity({ aid: command.aid, ...(command.ownerAid ? { ownerAid: command.ownerAid } : {}) })) };
+            }
+            catch {
+                return { text: "AUN 身份创建未能确认完成；身份可能已创建，请先查询 AID 列表确认状态，不要重复创建。" };
+            }
+        }
+        if (command?.kind === "bind") {
+            try {
+                return { text: safeJson(await management.bindAunIdentity({ aid: command.aid, agentId: command.agentId, replace: command.replace })) };
+            }
+            catch {
+                return { text: "AUN 身份绑定失败，请检查身份、目标和权限。" };
+            }
+        }
+        if (!this.client || this.state !== "ready")
+            return { text: "个人助理模型尚未配置，暂时无法处理请求。" };
+        const attachments = [];
+        const scopedManagement = { ...management, callAgent: async (id, task) => {
+                const result = await management.callAgent(id, task);
+                const items = result.attachments?.length ? result.attachments : (0, attachments_1.asAttachments)(result.images);
+                (0, attachments_1.validateAttachments)([...attachments, ...items]);
+                attachments.push(...items);
+                return result;
+            } };
+        const requestId = (0, node_crypto_1.randomUUID)();
+        const startedAt = Date.now();
+        const provider = this.context?.config.modelProvider === "plugin"
+            ? String(this.context.config.modelPlugin ?? "未指定插件")
+            : String(this.context?.config.modelProvider ?? "openai");
+        const sessionKey = (0, node_crypto_1.createHash)("sha256").update(`${this.manifest.id}\0${provider}\0${request.sessionId ?? (0, node_crypto_1.randomUUID)()}`).digest("hex");
+        try {
+            let context;
+            try {
+                const ownerContext = await this.ownerContext?.get();
+                if (ownerContext)
+                    context = `读取状态：${ownerContext.status}；缓存年龄：${ownerContext.ageMs} 毫秒。\n${ownerContext.content}`;
+            }
+            catch { /* 共享资料不可用时保留常规助理能力 */ }
+            const modelSessionId = sessionKey;
+            const result = await this.client.complete(request.task, scopedManagement, request.signal, requestTimeout(request.timeoutMs), context, modelSessionId);
+            this.context?.logger?.info("个人助理请求完成", { requestId, provider, durationMs: Date.now() - startedAt });
+            return { text: result, ...(attachments.length ? { attachments } : {}) };
+        }
+        catch (error) {
+            const detail = modelFailure(error, provider);
+            this.context?.logger?.warn("个人助理请求失败", { requestId, provider, session: sessionKey, stage: detail.stage, code: detail.code, durationMs: Date.now() - startedAt, ...(detail.message ? { error: detail.message } : {}) });
+            return { text: detail.text, ...(attachments.length ? { attachments } : {}) };
+        }
+    }
+}
+exports.PersonalAssistantPlugin = PersonalAssistantPlugin;
+function validEndpoint(value) {
+    if (typeof value !== "string" || !value.trim())
+        return undefined;
+    try {
+        const url = new URL(value);
+        if (!["http:", "https:"].includes(url.protocol) || url.username || url.password || url.hash || url.search || !url.pathname.endsWith("/chat/completions"))
+            return undefined;
+        return url.toString();
+    }
+    catch {
+        return undefined;
+    }
+}
+function isAid(value) { return typeof value === "string" && /^[A-Za-z0-9][A-Za-z0-9._-]{2,255}$/.test(value); }
+function isPluginId(value) { return typeof value === "string" && /^[a-z0-9][a-z0-9-]{0,63}$/.test(value); }
+function validTimeout(value) { return typeof value === "number" && Number.isInteger(value) && value >= 1000 && value <= 120_000 ? value : 120_000; }
+function requestTimeout(value) { return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : 120_000; }
+function validEcPath(value) { return typeof value === "string" && value.trim() ? value.trim() : undefined; }
+function validContextTtl(value) { return typeof value === "number" && Number.isSafeInteger(value) && value >= 1_000 && value <= 300_000 ? value : 60_000; }
+function safeJson(value) { try {
+    return JSON.stringify(value);
+}
+catch {
+    return "操作完成。";
+} }
+function modelFailure(error, provider) {
+    const operation = error instanceof agent_1.AgentOperationError ? error : undefined;
+    const stage = operation?.stage ?? "模型调用";
+    const code = operation?.code ?? "PA_REQUEST_FAILED";
+    if (code === "PA_REQUEST_CANCELLED")
+        return { text: `个人助理请求已取消（来源 ${provider}，阶段：${stage}）。`, stage, code };
+    if (code === "PA_REQUEST_TIMEOUT")
+        return { text: `个人助理请求超时（来源 ${provider}，阶段：${stage}）。`, stage, code };
+    const message = safeFailureMessage(operation?.cause ?? error);
+    return { text: `个人助理请求失败（来源 ${provider}，阶段：${stage}，错误码：${errorCode(error) ?? code}）${message ? `：${message}` : "。"}`, stage, code, ...(message ? { message } : {}) };
+}
+function errorCode(error) {
+    let result;
+    for (let current = error, count = 0; current && count++ < 5;) {
+        const code = current.code;
+        if (typeof code === "string" && /^[A-Z][A-Z0-9_]{2,31}$/.test(code))
+            result = code;
+        current = current.cause;
+    }
+    return result;
+}
+function safeFailureMessage(error) {
+    // 只回显宿主定义的诊断，插件任意异常可能含提示正文、凭据或本地路径。
+    const message = error instanceof Error ? error.message : "";
+    if (/^模型(?:服务请求失败：HTTP \d{3}|服务不允许重定向|响应超过大小限制|响应不是有效的工具决策 JSON|响应格式无效|未返回有效回复|返回了空回复|工具调用次数超过限制|返回的工具调用无效)$/.test(message))
+        return message;
+    if (/^(?:插件 [a-z0-9-]+ 未声明模型来源能力|模型来源插件 [a-z0-9-]+ 未安装)$/.test(message))
+        return message;
+    if (/^(?:Claude|Codex) CLI 模型请求(?:必须是包含 responseSchema 的 JSON 对象|缺少有效 responseSchema)$/.test(message))
+        return message;
+    const unavailable = /^模型来源插件“[^”\r\n]{1,64}”未就绪：([\s\S]+)$/.exec(message);
+    const reason = unavailable?.[1] ?? message;
+    if (/^(?:ready|degraded|failed|disabled|installed|discovered|starting|stopping|插件未实现模型来源接口)$/.test(reason))
+        return `模型来源未就绪：${reason}`;
+    const cdp = /(?:无法连接)?(千问办公|WorkBuddy|ChatGPT|豆包) CDP (?:端口 )?(\d{1,5})\b/.exec(reason);
+    if (cdp)
+        return `无法连接${cdp[1]} CDP 端口 ${cdp[2]}`;
+    const desktop = /(?:千问(?:办公)?|WorkBuddy|ChatGPT|豆包) ?(?:发送按钮(?:未就绪|不可用)|输入框(?:未就绪|未找到|含有未发送内容)|当前有未发送内容或正在生成回复，不能(?:新建|切换)会话)/.exec(reason);
+    if (desktop)
+        return desktop[0];
+    if (/^(?:WorkBuddy|ChatGPT|豆包) CDP (?:未连接|已断开)，模型请求不会启动或重启桌面$/.test(reason))
+        return reason;
+    if (reason === "千问办公未连接或不支持隔离模型会话，模型请求不会启动或重启桌面" || reason === "DSH 插件未就绪")
+        return reason;
+    if (reason === "WorkBuddy 当前有用户草稿、待确认操作或正在生成回复，模型请求未切换会话")
+        return reason;
+    const code = errorCode(error);
+    return code ? `底层错误码 ${code}` : undefined;
+}
+exports.default = PersonalAssistantPlugin;