@michengai/dsh-agency-agents 1.2.31 → 1.2.32

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,15 @@
1
1
  # Changelog
2
2
 
3
+ ## [1.2.32] - 2026-11-14
4
+
5
+ - **Configurable Brief Character Limit & Default Limit Raised**:
6
+ - **Raised Default Brief Budget**: Increased the default character limit for single expert tasks (`summon_expert` and `summon_experts`) from 2000 to 3000, providing generous headroom for complex delegations with extensive goals, constraints, file paths, and acceptance criteria;
7
+ - **Customizable Task Budget**: Added `summonTaskMaxChars` to plugin configuration, allowing users to customize the character budget (clamped to a safe `200–4000` range) with tool descriptions and validation fully synchronized;
8
+ - **Resolved Validation Guard Inconsistency**: Resolved an issue where modifying the character budget constant in bundle builds resulted in a mismatch between the advertised tool description and the active validation guard ([#39](https://github.com/MichengAI/dsh-agency-agents/issues/39)).
9
+ - **Compatibility & Dependency Updates**:
10
+ - Fully verified and adapted for the latest official DeepSeek Harness `0.2.2-rc.3` release; upgraded all development dependencies;
11
+ - Updated compatibility badges and verified version ranges across all 8 language manuals.
12
+
3
13
  ## [1.2.31] - 2026-11-13
4
14
 
5
15
  - **Unified settings header card**: wrapped the header in a bordered card container matching the modern ecosystem design.
@@ -1,5 +1,15 @@
1
1
  # 更新日志
2
2
 
3
+ ## [1.2.32] - 2026-11-14
4
+
5
+ - **专家召唤任务字数上限提升与支持配置**:
6
+ - **单条任务默认容量提升**:单条专家召唤(`summon_expert` 与 `summon_experts`)的任务描述默认字符上限由 2000 提升至 3000,为包含详细业务目标、多条边界约束、文件路径列表与验收标准的重度委派提供充裕空间;
7
+ - **支持自定义字数配置**:支持在插件配置中通过 `summonTaskMaxChars` 自由配置字数上限(提供 200–4000 的安全保护区间),模型工具描述与执行校验统一联动;
8
+ - **修复打包产物校验阈值割裂问题**:修复了在下游修改字数上限常量后工具提示与实际拦截阈值不一致的问题([#39](https://github.com/MichengAI/dsh-agency-agents/issues/39))。
9
+ - **兼容性与依赖升级**:
10
+ - 全面适配并验证官方最新发布宿主 DeepSeek Harness `0.2.2-rc.3`,开发依赖与类型定义同步更新;
11
+ - 8 门语言官方文档与徽章统一更新至 `0.2.2-rc.3`。
12
+
3
13
  ## [1.2.31] - 2026-11-13
4
14
 
5
15
  - **统一设置页头部卡片视觉**:顶部外层采用独立卡片边框容器,与全生态插件视觉风格完全对齐。
package/README.de.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![npm Downloads](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=downloads)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![DSH unterstützt bis 0.2.2--rc.2](https://img.shields.io/badge/DSH-0.2.2--rc.2-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
+ [![DSH unterstützt bis 0.2.2--rc.3](https://img.shields.io/badge/DSH-0.2.2--rc.3-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
21
21
 
22
22
  </div>
23
23
 
package/README.en.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![npm downloads](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=npm%20downloads)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![DSH supported through 0.2.2-rc.2](https://img.shields.io/badge/DSH-up%20to%200.2.2--rc.2-2563eb.svg)](#prerequisites)
20
+ [![DSH supported through 0.2.2-rc.3](https://img.shields.io/badge/DSH-up%20to%200.2.2--rc.3-2563eb.svg)](#prerequisites)
21
21
  </div>
22
22
 
23
23
  > DSH Agency Agents is a community-maintained DeepSeek Harness (DSH) plugin, not an official DeepSeek AI product.
@@ -102,7 +102,7 @@ Real-time expert and expert team suggestions displayed as a native dropdown at t
102
102
  - **One-Click Mention & Deduplication**: Click candidate items to insert references directly into the input via host `insertReference`; already referenced experts are automatically filtered out from suggestions, auto-dismissing on outside clicks or empty input;
103
103
  - **Dedicated Extension Toggle**: Toggle under **Settings → Plugins** (default off), supporting host volatile hot-reloading and persistence.
104
104
 
105
- > This README describes version 1.2.31. **This is a breaking upgrade: DeepSeek Harness 0.1.x is no longer supported.** Installation on an older host is rejected by the version gate. npm and GitHub Releases determine publication status. See the [changelog](CHANGELOG.md) for the full boundary. Expert and team settings are retained; back up settings before downgrading to 1.0.x.
105
+ > This README describes version 1.2.32. **This is a breaking upgrade: DeepSeek Harness 0.1.x is no longer supported.** Installation on an older host is rejected by the version gate. npm and GitHub Releases determine publication status. See the [changelog](CHANGELOG.md) for the full boundary. Expert and team settings are retained; back up settings before downgrading to 1.0.x.
106
106
  ## Screenshots
107
107
 
108
108
  The screenshots below show the 1.0.0 interface in Chinese. Experts and expert teams share the same settings page.
@@ -153,7 +153,7 @@ DSH `dsh.client.inject` controls browser module loading order. Missing graph ent
153
153
 
154
154
  ## Installation
155
155
 
156
- Host compatibility: Full DeepSeek Harness `0.2.x` line (`>=0.2.0-rc.1 <0.3.0-0`), verified with `0.2.0-rc.1`, `0.2.0-rc.2`, `0.2.1-rc.1`–`0.2.1-rc.5`, `0.2.2-rc.1`, and `0.2.2-rc.2`, and forwards-compatible with subsequent `0.2.x` RC releases without requiring routine plugin version bumps. Legacy versions and breaking changes in the `0.3.0` generation are rejected by the version gate. Expert and team settings are retained and do not need to be entered again. The host is supplied by your existing CLI or Desktop installation; the plugin does not install another copy.
156
+ Host compatibility: Full DeepSeek Harness `0.2.x` line (`>=0.2.0-rc.1 <0.3.0-0`), verified with `0.2.0-rc.1`, `0.2.0-rc.2`, `0.2.1-rc.1`–`0.2.1-rc.5`, `0.2.2-rc.1`, `0.2.2-rc.2`, and `0.2.2-rc.3`, and forwards-compatible with subsequent `0.2.x` RC releases without requiring routine plugin version bumps. Legacy versions and breaking changes in the `0.3.0` generation are rejected by the version gate. Expert and team settings are retained and do not need to be entered again. The host is supplied by your existing CLI or Desktop installation; the plugin does not install another copy.
157
157
 
158
158
  The installation commands below use the official npm registry.
159
159
 
@@ -809,6 +809,7 @@ Evaluating sources and synthesizing evidence into traceable findings.
809
809
  | `provider` | `spawn` | DSH subagent provider; `fork` also works when it supports persona and tool filtering. |
810
810
  | `divisions` | 22 standard divisions | Top-level divisions to scan. |
811
811
  | `maxDepth` | Unset | Positive absolute subagent-depth limit. |
812
+ | `summonTaskMaxChars` | `3000` | Per-brief Unicode character limit for domain expert summoning (clamped to 200–4000). |
812
813
 
813
814
  Set `AGENCY_AGENTS_ROOT` to use an external expert directory. Persona bodies from that directory are injected as the child system prompt, so load them only from a trusted source. The provider must support persona and tool filtering. `summon_experts` accepts at most 8 experts with a concurrency of 4 and still returns successful answers when some experts fail.
814
815
 
package/README.es.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![Descargas npm](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=downloads)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![DSH compatible hasta 0.2.2--rc.2](https://img.shields.io/badge/DSH-0.2.2--rc.2-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
+ [![DSH compatible hasta 0.2.2--rc.3](https://img.shields.io/badge/DSH-0.2.2--rc.3-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
21
21
 
22
22
  </div>
23
23
 
package/README.fr.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![Téléchargements npm](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=downloads)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![DSH supporté jusqu’à 0.2.2--rc.2](https://img.shields.io/badge/DSH-0.2.2--rc.2-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
+ [![DSH supporté jusqu’à 0.2.2--rc.3](https://img.shields.io/badge/DSH-0.2.2--rc.3-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
21
21
 
22
22
  </div>
23
23
 
package/README.ja.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![npm ダウンロード数](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=downloads)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![DSH サポート 0.2.2--rc.2 まで](https://img.shields.io/badge/DSH-0.2.2--rc.2-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
+ [![DSH サポート 0.2.2--rc.3 まで](https://img.shields.io/badge/DSH-0.2.2--rc.3-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
21
21
 
22
22
  </div>
23
23
 
package/README.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![npm 下载量](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=npm%20%E4%B8%8B%E8%BD%BD%E9%87%8F)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![DSH 支持至 0.2.2-rc.2](https://img.shields.io/badge/DSH-%E6%94%AF%E6%8C%81%E8%87%B3%200.2.2--rc.2-2563eb.svg)](#前置条件)
20
+ [![DSH 支持至 0.2.2-rc.3](https://img.shields.io/badge/DSH-%E6%94%AF%E6%8C%81%E8%87%B3%200.2.2--rc.3-2563eb.svg)](#前置条件)
21
21
  </div>
22
22
 
23
23
  > DSH Agency Agents 是社区维护的 DeepSeek Harness(DSH)插件,并非 DeepSeek AI 官方产品。
@@ -102,7 +102,7 @@
102
102
  - **一键引用与智能排除**:点击推荐项自动通过宿主原生 `insertReference` 将实体插入输入框草稿中,已引用的专家自动从推荐列表中排除,输入清空或点击外部时自动收起;
103
103
  - **独立配置开关**:在「设置 → 插件」扩展配置中提供「任务智能推荐」独立开关(默认关闭),支持实时热更新并持久化保存。
104
104
 
105
- > 本 README 描述 1.2.31。**这是破坏性升级:不再支持 DeepSeek Harness 0.1.x。** 在旧宿主上安装会被版本门拒绝。正式发行以 npm 和 GitHub Releases 为准。完整边界见[更新日志](CHANGELOG.zh-CN.md)。专家和专家团配置仍保留;回退 1.0.x 前先备份设置。
105
+ > 本 README 描述 1.2.32。**这是破坏性升级:不再支持 DeepSeek Harness 0.1.x。** 在旧宿主上安装会被版本门拒绝。正式发行以 npm 和 GitHub Releases 为准。完整边界见[更新日志](CHANGELOG.zh-CN.md)。专家和专家团配置仍保留;回退 1.0.x 前先备份设置。
106
106
  ## 界面预览
107
107
 
108
108
  以下为 1.0.0 中文界面截图,专家与专家团共用同一个设置页面。
@@ -153,7 +153,7 @@ DSH 的 `dsh.client.inject` 声明浏览器模块的加载顺序;宿主模块
153
153
 
154
154
  ## 安装
155
155
 
156
- 宿主支持范围:DeepSeek Harness `0.2.x` 全系列(`>=0.2.0-rc.1 <0.3.0-0`),已验证兼容 `0.2.0-rc.1`、`0.2.0-rc.2`、`0.2.1-rc.1`~`0.2.1-rc.5`、`0.2.2-rc.1`、`0.2.2-rc.2`,并自动支持后续全部 0.2.x RC 版本,同代宿主升级无需重复等待发版。早期版本与 `0.3.0` 代际破坏性更新由版本门拦截。专家和专家团配置仍保留,不需要重新录入。主程序由现有 CLI 或 Desktop 提供,插件不会另行安装。
156
+ 宿主支持范围:DeepSeek Harness `0.2.x` 全系列(`>=0.2.0-rc.1 <0.3.0-0`),已验证兼容 `0.2.0-rc.1`、`0.2.0-rc.2`、`0.2.1-rc.1`~`0.2.1-rc.5`、`0.2.2-rc.1`、`0.2.2-rc.2`、`0.2.2-rc.3`,并自动支持后续全部 0.2.x RC 版本,同代宿主升级无需重复等待发版。早期版本与 `0.3.0` 代际破坏性更新由版本门拦截。专家和专家团配置仍保留,不需要重新录入。主程序由现有 CLI 或 Desktop 提供,插件不会另行安装。
157
157
 
158
158
  以下安装命令使用官方 npm 源。
159
159
 
@@ -809,6 +809,7 @@ dsh --profile web --dump-config
809
809
  | `provider` | `spawn` | DSH 子代理 provider,可使用支持 persona 与工具过滤的 `fork`。 |
810
810
  | `divisions` | 22 个标准分区 | 需要扫描的顶层分区。 |
811
811
  | `maxDepth` | 未设置 | 正整数形式的绝对子代理深度上限。 |
812
+ | `summonTaskMaxChars` | `3000` | 单条专家召唤任务的 Unicode 字符上限(200–4000 范围钳制)。 |
812
813
 
813
814
  也可设置 `AGENCY_AGENTS_ROOT` 环境变量指定外部专家目录。外部目录里的 persona 正文会注入为子代理系统提示,只应从可信来源加载。子代理 provider 必须支持 persona 和工具过滤能力。`summon_experts` 一次最多 8 名专家,并发 4,部分失败仍返回成功结果。
814
815
 
package/README.pt.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![Transferências npm](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=downloads)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![DSH suportado até 0.2.2--rc.2](https://img.shields.io/badge/DSH-0.2.2--rc.2-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
+ [![DSH suportado até 0.2.2--rc.3](https://img.shields.io/badge/DSH-0.2.2--rc.3-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
21
21
 
22
22
  </div>
23
23
 
package/README.ru.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![Скачиваний npm](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=downloads)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![Поддержка DSH до 0.2.2--rc.2](https://img.shields.io/badge/DSH-0.2.2--rc.2-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
+ [![Поддержка DSH до 0.2.2--rc.3](https://img.shields.io/badge/DSH-0.2.2--rc.3-2563eb.svg)](https://github.com/MichengAI/dsh-agency-agents)
21
21
 
22
22
  </div>
23
23
 
package/README.zh-CN.md CHANGED
@@ -17,7 +17,7 @@
17
17
  [![npm package](https://img.shields.io/npm/v/%40michengai%2Fdsh-agency-agents.svg?label=npm%20package)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
18
18
  [![npm 下载量](https://img.shields.io/npm/dt/%40michengai%2Fdsh-agency-agents.svg?label=npm%20%E4%B8%8B%E8%BD%BD%E9%87%8F)](https://www.npmjs.com/package/@michengai/dsh-agency-agents)
19
19
  [![DSH Web Plugin](https://img.shields.io/badge/DSH%20Web-Plugin-0f766e.svg)](https://github.com/MichengAI/dsh-agency-agents)
20
- [![DSH 支持至 0.2.2-rc.2](https://img.shields.io/badge/DSH-%E6%94%AF%E6%8C%81%E8%87%B3%200.2.2--rc.2-2563eb.svg)](#前置条件)
20
+ [![DSH 支持至 0.2.2-rc.3](https://img.shields.io/badge/DSH-%E6%94%AF%E6%8C%81%E8%87%B3%200.2.2--rc.3-2563eb.svg)](#前置条件)
21
21
  </div>
22
22
 
23
23
  > DSH Agency Agents 是社区维护的 DeepSeek Harness(DSH)插件,并非 DeepSeek AI 官方产品。
@@ -102,7 +102,7 @@
102
102
  - **一键引用与智能排除**:点击推荐项自动通过宿主原生 `insertReference` 将实体插入输入框草稿中,已引用的专家自动从推荐列表中排除,输入清空或点击外部时自动收起;
103
103
  - **独立配置开关**:在「设置 → 插件」扩展配置中提供「任务智能推荐」独立开关(默认关闭),支持实时热更新并持久化保存。
104
104
 
105
- > 本 README 描述 1.2.31。**这是破坏性升级:不再支持 DeepSeek Harness 0.1.x。** 在旧宿主上安装会被版本门拒绝。正式发行以 npm 和 GitHub Releases 为准。完整边界见[更新日志](CHANGELOG.zh-CN.md)。专家和专家团配置仍保留;回退 1.0.x 前先备份设置。
105
+ > 本 README 描述 1.2.32。**这是破坏性升级:不再支持 DeepSeek Harness 0.1.x。** 在旧宿主上安装会被版本门拒绝。正式发行以 npm 和 GitHub Releases 为准。完整边界见[更新日志](CHANGELOG.zh-CN.md)。专家和专家团配置仍保留;回退 1.0.x 前先备份设置。
106
106
  ## 界面预览
107
107
 
108
108
  以下为 1.0.0 中文界面截图,专家与专家团共用同一个设置页面。
@@ -153,7 +153,7 @@ DSH 的 `dsh.client.inject` 声明浏览器模块的加载顺序;宿主模块
153
153
 
154
154
  ## 安装
155
155
 
156
- 宿主支持范围:DeepSeek Harness `0.2.x` 全系列(`>=0.2.0-rc.1 <0.3.0-0`),已验证兼容 `0.2.0-rc.1`、`0.2.0-rc.2`、`0.2.1-rc.1`~`0.2.1-rc.5`、`0.2.2-rc.1`、`0.2.2-rc.2`,并自动支持后续全部 0.2.x RC 版本,同代宿主升级无需重复等待发版。早期版本与 `0.3.0` 代际破坏性更新由版本门拦截。专家和专家团配置仍保留,不需要重新录入。主程序由现有 CLI 或 Desktop 提供,插件不会另行安装。
156
+ 宿主支持范围:DeepSeek Harness `0.2.x` 全系列(`>=0.2.0-rc.1 <0.3.0-0`),已验证兼容 `0.2.0-rc.1`、`0.2.0-rc.2`、`0.2.1-rc.1`~`0.2.1-rc.5`、`0.2.2-rc.1`、`0.2.2-rc.2`、`0.2.2-rc.3`,并自动支持后续全部 0.2.x RC 版本,同代宿主升级无需重复等待发版。早期版本与 `0.3.0` 代际破坏性更新由版本门拦截。专家和专家团配置仍保留,不需要重新录入。主程序由现有 CLI 或 Desktop 提供,插件不会另行安装。
157
157
 
158
158
  以下安装命令使用官方 npm 源。
159
159
 
@@ -809,6 +809,7 @@ dsh --profile web --dump-config
809
809
  | `provider` | `spawn` | DSH 子代理 provider,可使用支持 persona 与工具过滤的 `fork`。 |
810
810
  | `divisions` | 22 个标准分区 | 需要扫描的顶层分区。 |
811
811
  | `maxDepth` | 未设置 | 正整数形式的绝对子代理深度上限。 |
812
+ | `summonTaskMaxChars` | `3000` | 单条专家召唤任务的 Unicode 字符上限(200–4000 范围钳制)。 |
812
813
 
813
814
  也可设置 `AGENCY_AGENTS_ROOT` 环境变量指定外部专家目录。外部目录里的 persona 正文会注入为子代理系统提示,只应从可信来源加载。子代理 provider 必须支持 persona 和工具过滤能力。`summon_experts` 一次最多 8 名专家,并发 4,部分失败仍返回成功结果。
814
815
 
@@ -17,8 +17,12 @@ declare const DEFAULT_DIVISIONS: string[];
17
17
  declare const SUMMON_EXPERTS_MAX = 8;
18
18
  /** 批量召唤的并发上限。 */
19
19
  declare const SUMMON_EXPERTS_CONCURRENCY = 4;
20
- /** 单条任务的 Unicode 码点上限。本机实测单条约 3200 字时工具 JSON 仍合法,约 7200 字时未转义引号会让宿主在专家启动前结束回合。2000 是给目标、约束和路径留的余量,不是算出的安全边界。 */
21
- declare const SUMMON_TASK_MAX_CHARS = 2000;
20
+ /** 单条任务的 Unicode 码点上限。默认 3000。本机实测单条约 3200 字时工具 JSON 仍合法,约 7200 字时未转义引号会让宿主在专家启动前结束回合。 */
21
+ declare const SUMMON_TASK_MAX_CHARS = 3000;
22
+ declare const SUMMON_TASK_MIN_LIMIT = 200;
23
+ declare const SUMMON_TASK_MAX_LIMIT = 4000;
24
+ /** 规范化单条任务字符上限:缺省使用默认 3000,越界值钳制在 200–4000 安全区间。 */
25
+ declare function normalizeSummonTaskMaxChars(value: unknown): number;
22
26
  /** 已校验的批量召唤条目。 */
23
27
  interface SummonExpertSpec {
24
28
  readonly expert: unknown;
@@ -32,7 +36,7 @@ interface SummonExpertItemResult {
32
36
  readonly error?: string;
33
37
  }
34
38
  /** 校验批量召唤入参:非空、数量上限、专家名非空、任务非空且不超过码点上限。 */
35
- declare function validateSummonSpecs(specs: unknown, locale: LocaleId): SummonExpertSpec[];
39
+ declare function validateSummonSpecs(specs: unknown, locale: LocaleId, maxChars?: number): SummonExpertSpec[];
36
40
  /** 受限并发地映射异步任务,结果顺序与输入一致。 */
37
41
  declare function mapPool<T, R>(items: readonly T[], concurrency: number, mapper: (item: T, index: number) => Promise<R>): Promise<R[]>;
38
42
  /** 把单次专家运行结果收成批量条目;失败时保留原始查询作为专家名。 */
@@ -62,6 +66,8 @@ interface Config {
62
66
  /** 可选的正整数绝对子代理深度上限;未设置时沿用 provider 的默认行为。 */
63
67
  maxDepth?: number;
64
68
  sidebarEntry?: boolean;
69
+ /** 单条专家召唤任务的 Unicode 码点上限。默认 3000,超出 200–4000 安全区间时会被自动钳制。 */
70
+ summonTaskMaxChars?: number;
65
71
  }
66
72
  declare const AGENCY_PERSONA_SERVICE = "agencyAgentsPersona";
67
73
  /** 仅当宿主 schemastery 提供 volatile() 时,把字段标成可实时更新。插件自己的新版本不能代替宿主判断。 */
@@ -111,4 +117,4 @@ declare function resolveExpert<T extends {
111
117
  }>(experts: readonly T[], query: unknown, locale?: LocaleId): T;
112
118
  declare function apply(ctx: Context, config: Config): void;
113
119
  //#endregion
114
- export { stripBom as C, validateSummonSpecs as D, unquote as E, ZH_NAME as O, sanitize as S, truncate as T, parseFrontmatter as _, SUMMON_EXPERTS_CONCURRENCY as a, resolveCatalogRoot as b, SummonExpertItemResult as c, createAgencyPersonaSource as d, inject as f, name as g, mapPool as h, DEFAULT_DIVISIONS as i, SummonExpertSpec as l, loadCatalog as m, AgencyPersonaSource as n, SUMMON_EXPERTS_MAX as o, liveSchemaField as p, Config as r, SUMMON_TASK_MAX_CHARS as s, AGENCY_PERSONA_SERVICE as t, apply as u, readExpertPrompt as v, toSummonItemResult as w, resolveExpert as x, readLocalizedExpertPrompt as y };
120
+ export { validateSummonSpecs as A, resolveCatalogRoot as C, toSummonItemResult as D, stripBom as E, truncate as O, readLocalizedExpertPrompt as S, sanitize as T, mapPool as _, SUMMON_EXPERTS_CONCURRENCY as a, parseFrontmatter as b, SUMMON_TASK_MAX_LIMIT as c, SummonExpertSpec as d, apply as f, loadCatalog as g, liveSchemaField as h, DEFAULT_DIVISIONS as i, ZH_NAME as j, unquote as k, SUMMON_TASK_MIN_LIMIT as l, inject as m, AgencyPersonaSource as n, SUMMON_EXPERTS_MAX as o, createAgencyPersonaSource as p, Config as r, SUMMON_TASK_MAX_CHARS as s, AGENCY_PERSONA_SERVICE as t, SummonExpertItemResult as u, name as v, resolveExpert as w, readExpertPrompt as x, normalizeSummonTaskMaxChars as y };
package/lib/index.d.ts CHANGED
@@ -1,2 +1,2 @@
1
- import { C as stripBom, D as validateSummonSpecs, E as unquote, O as ZH_NAME, S as sanitize, T as truncate, _ as parseFrontmatter, a as SUMMON_EXPERTS_CONCURRENCY, b as resolveCatalogRoot, c as SummonExpertItemResult, d as createAgencyPersonaSource, f as inject, g as name, h as mapPool, i as DEFAULT_DIVISIONS, l as SummonExpertSpec, m as loadCatalog, n as AgencyPersonaSource, o as SUMMON_EXPERTS_MAX, p as liveSchemaField, r as Config, s as SUMMON_TASK_MAX_CHARS, t as AGENCY_PERSONA_SERVICE, u as apply, v as readExpertPrompt, w as toSummonItemResult, x as resolveExpert, y as readLocalizedExpertPrompt } from "./index-DpzRvHVP.js";
2
- export { AGENCY_PERSONA_SERVICE, AgencyPersonaSource, Config, DEFAULT_DIVISIONS, SUMMON_EXPERTS_CONCURRENCY, SUMMON_EXPERTS_MAX, SUMMON_TASK_MAX_CHARS, SummonExpertItemResult, SummonExpertSpec, ZH_NAME, apply, createAgencyPersonaSource, inject, liveSchemaField, loadCatalog, mapPool, name, parseFrontmatter, readExpertPrompt, readLocalizedExpertPrompt, resolveCatalogRoot, resolveExpert, sanitize, stripBom, toSummonItemResult, truncate, unquote, validateSummonSpecs };
1
+ import { A as validateSummonSpecs, C as resolveCatalogRoot, D as toSummonItemResult, E as stripBom, O as truncate, S as readLocalizedExpertPrompt, T as sanitize, _ as mapPool, a as SUMMON_EXPERTS_CONCURRENCY, b as parseFrontmatter, c as SUMMON_TASK_MAX_LIMIT, d as SummonExpertSpec, f as apply, g as loadCatalog, h as liveSchemaField, i as DEFAULT_DIVISIONS, j as ZH_NAME, k as unquote, l as SUMMON_TASK_MIN_LIMIT, m as inject, n as AgencyPersonaSource, o as SUMMON_EXPERTS_MAX, p as createAgencyPersonaSource, r as Config, s as SUMMON_TASK_MAX_CHARS, t as AGENCY_PERSONA_SERVICE, u as SummonExpertItemResult, v as name, w as resolveExpert, x as readExpertPrompt, y as normalizeSummonTaskMaxChars } from "./index-BaN3r1Yj.js";
2
+ export { AGENCY_PERSONA_SERVICE, AgencyPersonaSource, Config, DEFAULT_DIVISIONS, SUMMON_EXPERTS_CONCURRENCY, SUMMON_EXPERTS_MAX, SUMMON_TASK_MAX_CHARS, SUMMON_TASK_MAX_LIMIT, SUMMON_TASK_MIN_LIMIT, SummonExpertItemResult, SummonExpertSpec, ZH_NAME, apply, createAgencyPersonaSource, inject, liveSchemaField, loadCatalog, mapPool, name, normalizeSummonTaskMaxChars, parseFrontmatter, readExpertPrompt, readLocalizedExpertPrompt, resolveCatalogRoot, resolveExpert, sanitize, stripBom, toSummonItemResult, truncate, unquote, validateSummonSpecs };
package/lib/index.js CHANGED
@@ -1,2 +1,2 @@
1
- import { C as unquote, I as ZH_NAME, S as truncate, _ as resolveCatalogRoot, a as SUMMON_EXPERTS_MAX, b as stripBom, c as createAgencyPersonaSource, d as loadCatalog, f as mapPool, g as readLocalizedExpertPrompt, h as readExpertPrompt, i as SUMMON_EXPERTS_CONCURRENCY, l as inject, m as parseFrontmatter, n as Config, o as SUMMON_TASK_MAX_CHARS, p as name, r as DEFAULT_DIVISIONS, s as apply, t as AGENCY_PERSONA_SERVICE, u as liveSchemaField, v as resolveExpert, w as validateSummonSpecs, x as toSummonItemResult, y as sanitize } from "./src-CRp2hHJs.js";
2
- export { AGENCY_PERSONA_SERVICE, Config, DEFAULT_DIVISIONS, SUMMON_EXPERTS_CONCURRENCY, SUMMON_EXPERTS_MAX, SUMMON_TASK_MAX_CHARS, ZH_NAME, apply, createAgencyPersonaSource, inject, liveSchemaField, loadCatalog, mapPool, name, parseFrontmatter, readExpertPrompt, readLocalizedExpertPrompt, resolveCatalogRoot, resolveExpert, sanitize, stripBom, toSummonItemResult, truncate, unquote, validateSummonSpecs };
1
+ import { C as stripBom, D as validateSummonSpecs, E as unquote, S as sanitize, T as truncate, _ as parseFrontmatter, a as SUMMON_EXPERTS_MAX, b as resolveCatalogRoot, c as SUMMON_TASK_MIN_LIMIT, d as inject, f as liveSchemaField, g as normalizeSummonTaskMaxChars, h as name, i as SUMMON_EXPERTS_CONCURRENCY, l as apply, m as mapPool, n as Config, o as SUMMON_TASK_MAX_CHARS, p as loadCatalog, r as DEFAULT_DIVISIONS, s as SUMMON_TASK_MAX_LIMIT, t as AGENCY_PERSONA_SERVICE, u as createAgencyPersonaSource, v as readExpertPrompt, w as toSummonItemResult, x as resolveExpert, y as readLocalizedExpertPrompt, z as ZH_NAME } from "./src-B-RkXN6b.js";
2
+ export { AGENCY_PERSONA_SERVICE, Config, DEFAULT_DIVISIONS, SUMMON_EXPERTS_CONCURRENCY, SUMMON_EXPERTS_MAX, SUMMON_TASK_MAX_CHARS, SUMMON_TASK_MAX_LIMIT, SUMMON_TASK_MIN_LIMIT, ZH_NAME, apply, createAgencyPersonaSource, inject, liveSchemaField, loadCatalog, mapPool, name, normalizeSummonTaskMaxChars, parseFrontmatter, readExpertPrompt, readLocalizedExpertPrompt, resolveCatalogRoot, resolveExpert, sanitize, stripBom, toSummonItemResult, truncate, unquote, validateSummonSpecs };
package/lib/remote.d.ts CHANGED
@@ -1,4 +1,4 @@
1
- import { v as readExpertPrompt, y as readLocalizedExpertPrompt } from "./index-DpzRvHVP.js";
1
+ import { S as readLocalizedExpertPrompt, x as readExpertPrompt } from "./index-BaN3r1Yj.js";
2
2
  import { z } from "zod";
3
3
  import { TypertRemoteService } from "@deepseek-ai/dsh-typert-protocol";
4
4
  import { Context } from "@deepseek-ai/cordis";
package/lib/remote.js CHANGED
@@ -1,4 +1,4 @@
1
- import { A as settingsNamespaceCompat, B as teamText, D as storedProposalSchema, E as latestProposalSchema, F as expertEditSchema, L as AGENCY_TEAM_SERVICE, M as CUSTOM_EXPERT_SLUG, N as catalogSnapshotSchema, O as formatHost, P as customExpertInputSchema, R as teamInputSchema, T as nativeTeamMemberName, g as readLocalizedExpertPrompt, h as readExpertPrompt, j as AGENCY_LIBRARY_SERVICE, k as readHostLocale, t as AGENCY_PERSONA_SERVICE, z as teamSnapshotSchema } from "./src-CRp2hHJs.js";
1
+ import { A as storedProposalSchema, B as AGENCY_TEAM_SERVICE, F as CUSTOM_EXPERT_SLUG, H as teamSnapshotSchema, I as catalogSnapshotSchema, L as customExpertInputSchema, M as readHostLocale, N as settingsNamespaceCompat, O as nativeTeamMemberName, P as AGENCY_LIBRARY_SERVICE, R as expertEditSchema, U as teamText, V as teamInputSchema, j as formatHost, k as latestProposalSchema, t as AGENCY_PERSONA_SERVICE, v as readExpertPrompt, y as readLocalizedExpertPrompt } from "./src-B-RkXN6b.js";
2
2
  import { z } from "zod";
3
3
  import { Remote, TypertRemoteService } from "@deepseek-ai/dsh-typert-protocol";
4
4
  //#region src/remote-contract.ts
@@ -3135,33 +3135,41 @@ const DESCRIPTION_LIMIT = 120;
3135
3135
  const SUMMON_EXPERTS_MAX = 8;
3136
3136
  /** 批量召唤的并发上限。 */
3137
3137
  const SUMMON_EXPERTS_CONCURRENCY = 4;
3138
- /** 单条任务的 Unicode 码点上限。本机实测单条约 3200 字时工具 JSON 仍合法,约 7200 字时未转义引号会让宿主在专家启动前结束回合。2000 是给目标、约束和路径留的余量,不是算出的安全边界。 */
3139
- const SUMMON_TASK_MAX_CHARS = 2e3;
3138
+ /** 单条任务的 Unicode 码点上限。默认 3000。本机实测单条约 3200 字时工具 JSON 仍合法,约 7200 字时未转义引号会让宿主在专家启动前结束回合。 */
3139
+ const SUMMON_TASK_MAX_CHARS = 3e3;
3140
+ const SUMMON_TASK_MIN_LIMIT = 200;
3141
+ const SUMMON_TASK_MAX_LIMIT = 4e3;
3142
+ /** 规范化单条任务字符上限:缺省使用默认 3000,越界值钳制在 200–4000 安全区间。 */
3143
+ function normalizeSummonTaskMaxChars(value) {
3144
+ if (value === void 0 || value === null) return SUMMON_TASK_MAX_CHARS;
3145
+ if (typeof value !== "number" || !Number.isFinite(value)) return SUMMON_TASK_MAX_CHARS;
3146
+ return Math.max(200, Math.min(SUMMON_TASK_MAX_LIMIT, Math.round(value)));
3147
+ }
3140
3148
  /** 召唤参数说明。模型把带引号的源码写进工具 JSON 时,宿主会在执行前拒绝整轮。 */
3141
- function summonTaskDescription(scope) {
3142
- const shared = `Short brief only: the user goal, constraints, and file paths. Do not paste source, CSS, HTML, or quoted code; an unescaped quote makes the tool JSON invalid and no expert starts. At most ${SUMMON_TASK_MAX_CHARS} characters. The expert reads the files itself.`;
3149
+ function summonTaskDescription(scope, maxChars = SUMMON_TASK_MAX_CHARS) {
3150
+ const shared = `Short brief only: the user goal, constraints, and file paths. Do not paste source, CSS, HTML, or quoted code; an unescaped quote makes the tool JSON invalid and no expert starts. At most ${maxChars} characters. The expert reads the files itself.`;
3143
3151
  return scope === "member" ? `${shared} One sentence of this expert's role is enough; do not repeat a long brief for every expert.` : shared;
3144
3152
  }
3145
3153
  /**
3146
3154
  * 校验并规范化任务文本:非空且不超过码点上限。
3147
3155
  * index 存在时使用带序号的批量文案,否则使用单条召唤文案;返回规范化后的字符串。
3148
3156
  */
3149
- function normalizeTask(task, locale, index) {
3157
+ function normalizeTask(task, locale, index, maxChars = SUMMON_TASK_MAX_CHARS) {
3150
3158
  const text = task === void 0 || task === null ? "" : String(task);
3151
3159
  const length = Array.from(text).length;
3152
3160
  if (text.trim() === "") throw new Error(index === void 0 ? formatHost(locale, "error.taskRequired") : formatHost(locale, "error.taskEmpty", { index }));
3153
- if (length > 2e3) throw new Error(index === void 0 ? formatHost(locale, "error.taskLimit", {
3161
+ if (length > maxChars) throw new Error(index === void 0 ? formatHost(locale, "error.taskLimit", {
3154
3162
  length,
3155
- max: SUMMON_TASK_MAX_CHARS
3163
+ max: maxChars
3156
3164
  }) : formatHost(locale, "error.taskTooLong", {
3157
3165
  index,
3158
3166
  length,
3159
- max: SUMMON_TASK_MAX_CHARS
3167
+ max: maxChars
3160
3168
  }));
3161
3169
  return text;
3162
3170
  }
3163
3171
  /** 校验批量召唤入参:非空、数量上限、专家名非空、任务非空且不超过码点上限。 */
3164
- function validateSummonSpecs(specs, locale) {
3172
+ function validateSummonSpecs(specs, locale, maxChars = SUMMON_TASK_MAX_CHARS) {
3165
3173
  if (!Array.isArray(specs) || specs.length === 0) throw new Error(formatHost(locale, "error.expertsEmpty"));
3166
3174
  if (specs.length > 8) throw new Error(formatHost(locale, "error.expertsTooMany", {
3167
3175
  max: 8,
@@ -3173,7 +3181,7 @@ function validateSummonSpecs(specs, locale) {
3173
3181
  if (expert === void 0 || expert === null || String(expert).trim() === "") throw new Error(formatHost(locale, "error.expertEmpty", { index: index + 1 }));
3174
3182
  return {
3175
3183
  expert,
3176
- task: normalizeTask(record?.task, locale, index + 1)
3184
+ task: normalizeTask(record?.task, locale, index + 1, maxChars)
3177
3185
  };
3178
3186
  });
3179
3187
  }
@@ -3237,6 +3245,7 @@ const configFields = {
3237
3245
  provider: schema$1.string().default("spawn"),
3238
3246
  divisions: schema$1.array(schema$1.string()).default(DEFAULT_DIVISIONS),
3239
3247
  maxDepth: schema$1.natural().min(1),
3248
+ summonTaskMaxChars: schema$1.natural().min(200).max(SUMMON_TASK_MAX_LIMIT).default(SUMMON_TASK_MAX_CHARS),
3240
3249
  entryLocation: schema$1.string(),
3241
3250
  sidebarEntry: schema$1.boolean(),
3242
3251
  composerRecommend: schema$1.boolean()
@@ -3248,7 +3257,8 @@ for (const [key, field] of [
3248
3257
  ["enabledTeams", schema$1.array(schema$1.string()).default([])],
3249
3258
  ["entryLocation", schema$1.string()],
3250
3259
  ["sidebarEntry", schema$1.boolean()],
3251
- ["composerRecommend", schema$1.boolean()]
3260
+ ["composerRecommend", schema$1.boolean()],
3261
+ ["summonTaskMaxChars", schema$1.natural().min(200).max(SUMMON_TASK_MAX_LIMIT).default(SUMMON_TASK_MAX_CHARS)]
3252
3262
  ]) {
3253
3263
  const live = markLiveFields ? liveSchemaField(field) : void 0;
3254
3264
  if (live !== void 0) configFields[key] = live;
@@ -3544,6 +3554,7 @@ async function readJsonBody(request) {
3544
3554
  }
3545
3555
  function apply(ctx, config) {
3546
3556
  const maxDepth = normalizeMaxDepth(config.maxDepth);
3557
+ const getTaskMaxChars = () => normalizeSummonTaskMaxChars(config.summonTaskMaxChars);
3547
3558
  const settingsNamespace = settingsNamespaceCompat("agency-agents");
3548
3559
  let settingsSource = () => ({
3549
3560
  enabled: [],
@@ -3996,7 +4007,7 @@ function apply(ctx, config) {
3996
4007
  }
3997
4008
  async function runExpert(query, task, exec) {
3998
4009
  const locale = activeLocale();
3999
- const taskText = normalizeTask(task, locale);
4010
+ const taskText = normalizeTask(task, locale, void 0, getTaskMaxChars());
4000
4011
  if (exec.agent === void 0) throw new Error(formatHost(locale, "error.summonRequiresAgent"));
4001
4012
  const provider = ctx.subagents.getProvider(config.provider);
4002
4013
  if (provider === void 0) throw new Error(formatHost(locale, "error.providerMissing", { provider: config.provider }));
@@ -4047,7 +4058,7 @@ function apply(ctx, config) {
4047
4058
  task: {
4048
4059
  type: "string",
4049
4060
  required: true,
4050
- description: summonTaskDescription("single")
4061
+ description: summonTaskDescription("single", getTaskMaxChars())
4051
4062
  }
4052
4063
  },
4053
4064
  output: {
@@ -4094,7 +4105,7 @@ function apply(ctx, config) {
4094
4105
  task: {
4095
4106
  type: "string",
4096
4107
  required: true,
4097
- description: summonTaskDescription("member")
4108
+ description: summonTaskDescription("member", getTaskMaxChars())
4098
4109
  }
4099
4110
  }
4100
4111
  }
@@ -4121,7 +4132,7 @@ function apply(ctx, config) {
4121
4132
  await ensureReady();
4122
4133
  const locale = activeLocale();
4123
4134
  if (exec.agent === void 0) throw new Error(formatHost(locale, "error.summonManyRequiresAgent"));
4124
- return { results: (await mapPool(validateSummonSpecs(args.experts, locale), 4, async (spec) => {
4135
+ return { results: (await mapPool(validateSummonSpecs(args.experts, locale, getTaskMaxChars()), 4, async (spec) => {
4125
4136
  try {
4126
4137
  return toSummonItemResult(spec.expert, await runExpert(spec.expert, spec.task, exec));
4127
4138
  } catch (error) {
@@ -4619,9 +4630,9 @@ function apply(ctx, config) {
4619
4630
  order: 117,
4620
4631
  text: (context) => {
4621
4632
  if (context.agent?.session?.header?.parentSession !== void 0) return "";
4622
- return "## Agency expert mode\nThe parent session has a roster of domain experts from The Agency (specialists across 22 divisions, individually enable/disable; ALL are disabled by default, and the user enables some in the Agency settings tab). A composer selection inserts one enabled expert as a native reference chip; all remaining draft text is that expert's task. When the user message already names experts, call summon_expert or summon_experts immediately. Summarize that message into the user goal, constraints, and file paths; do not pass the raw text when it is long. Do not call list_experts, read the project, or do the analysis yourself. Experts read the files and produce the analysis; the parent only synthesizes their results. Call list_experts only when the user did not name an expert. summon_experts runs at most 8 experts; partial results are returned if some fail. A disabled expert cannot be summoned. When calling either tool, pass only a short brief: the user goal, constraints, and file paths. Do not paste source, CSS, HTML, or quoted code into the tool argument; an unescaped quote makes the JSON invalid and no expert starts. Each task must stay within 2000 characters. Experts read the files themselves.\nWhen the user asks to create, design, or customize an expert persona (e.g. \"我想创建一个专家...\", \"创建小红书运营专家\", \"新建专家\"), call propose_custom_expert. You MUST ALWAYS present a full, polished preview in your response text (name, division, description, and detailed system prompt breakdown) so the user can read and review it first. Invite the user to suggest modifications in chat. Only when the user explicitly confirms (e.g. says \"确认\", \"就按这个\", \"打开弹窗\", \"满意\"), call open_custom_expert_editor to open the modal. Do NOT tell the user to manually copy-paste, and do NOT write markdown files to docs/. When the user asks to assemble, design, or create an expert team (e.g. \"帮我组建一个专家团队...\", \"创建专家团\"), call propose_expert_team. Always present team members using their localized names (e.g. 中文名如「产品经理」、「代码审查工程师」) alongside their slug, duties, and reasons in your chat preview. Do NOT tell the user to click non-existent links; instead, inform them that they can review the draft and say \"确认\" when satisfied. When the user explicitly confirms (e.g. says \"确认\", \"就按这个\", \"满意\", \"打开弹窗\"), call open_expert_team_editor (or open_custom_expert_editor), which will present an interactive confirmation card before popping up the full editor dialog.";
4633
+ return "## Agency expert mode\nThe parent session has a roster of domain experts from The Agency (specialists across 22 divisions, individually enable/disable; ALL are disabled by default, and the user enables some in the Agency settings tab). A composer selection inserts one enabled expert as a native reference chip; all remaining draft text is that expert's task. When the user message already names experts, call summon_expert or summon_experts immediately. Summarize that message into the user goal, constraints, and file paths; do not pass the raw text when it is long. Do not call list_experts, read the project, or do the analysis yourself. Experts read the files and produce the analysis; the parent only synthesizes their results. Call list_experts only when the user did not name an expert. summon_experts runs at most 8 experts; partial results are returned if some fail. A disabled expert cannot be summoned. When calling either tool, pass only a short brief: the user goal, constraints, and file paths. Do not paste source, CSS, HTML, or quoted code into the tool argument; an unescaped quote makes the JSON invalid and no expert starts. Each task must stay within " + getTaskMaxChars() + " characters. Experts read the files themselves.\nWhen the user asks to create, design, or customize an expert persona (e.g. \"我想创建一个专家...\", \"创建小红书运营专家\", \"新建专家\"), call propose_custom_expert. You MUST ALWAYS present a full, polished preview in your response text (name, division, description, and detailed system prompt breakdown) so the user can read and review it first. Invite the user to suggest modifications in chat. Only when the user explicitly confirms (e.g. says \"确认\", \"就按这个\", \"打开弹窗\", \"满意\"), call open_custom_expert_editor to open the modal. Do NOT tell the user to manually copy-paste, and do NOT write markdown files to docs/. When the user asks to assemble, design, or create an expert team (e.g. \"帮我组建一个专家团队...\", \"创建专家团\"), call propose_expert_team. Always present team members using their localized names (e.g. 中文名如「产品经理」、「代码审查工程师」) alongside their slug, duties, and reasons in your chat preview. Do NOT tell the user to click non-existent links; instead, inform them that they can review the draft and say \"确认\" when satisfied. When the user explicitly confirms (e.g. says \"确认\", \"就按这个\", \"满意\", \"打开弹窗\"), call open_expert_team_editor (or open_custom_expert_editor), which will present an interactive confirmation card before popping up the full editor dialog.";
4623
4634
  }
4624
4635
  });
4625
4636
  }
4626
4637
  //#endregion
4627
- export { settingsNamespaceCompat as A, teamText as B, unquote as C, storedProposalSchema as D, latestProposalSchema as E, expertEditSchema as F, ZH_NAME as I, AGENCY_TEAM_SERVICE as L, CUSTOM_EXPERT_SLUG as M, catalogSnapshotSchema as N, formatHost as O, customExpertInputSchema as P, teamInputSchema as R, truncate as S, nativeTeamMemberName as T, resolveCatalogRoot as _, SUMMON_EXPERTS_MAX as a, stripBom as b, createAgencyPersonaSource as c, loadCatalog as d, mapPool as f, readLocalizedExpertPrompt as g, readExpertPrompt as h, SUMMON_EXPERTS_CONCURRENCY as i, AGENCY_LIBRARY_SERVICE as j, readHostLocale as k, inject as l, parseFrontmatter as m, Config as n, SUMMON_TASK_MAX_CHARS as o, name as p, DEFAULT_DIVISIONS as r, apply as s, AGENCY_PERSONA_SERVICE as t, liveSchemaField as u, resolveExpert as v, validateSummonSpecs as w, toSummonItemResult as x, sanitize as y, teamSnapshotSchema as z };
4638
+ export { storedProposalSchema as A, AGENCY_TEAM_SERVICE as B, stripBom as C, validateSummonSpecs as D, unquote as E, CUSTOM_EXPERT_SLUG as F, teamSnapshotSchema as H, catalogSnapshotSchema as I, customExpertInputSchema as L, readHostLocale as M, settingsNamespaceCompat as N, nativeTeamMemberName as O, AGENCY_LIBRARY_SERVICE as P, expertEditSchema as R, sanitize as S, truncate as T, teamText as U, teamInputSchema as V, parseFrontmatter as _, SUMMON_EXPERTS_MAX as a, resolveCatalogRoot as b, SUMMON_TASK_MIN_LIMIT as c, inject as d, liveSchemaField as f, normalizeSummonTaskMaxChars as g, name as h, SUMMON_EXPERTS_CONCURRENCY as i, formatHost as j, latestProposalSchema as k, apply as l, mapPool as m, Config as n, SUMMON_TASK_MAX_CHARS as o, loadCatalog as p, DEFAULT_DIVISIONS as r, SUMMON_TASK_MAX_LIMIT as s, AGENCY_PERSONA_SERVICE as t, createAgencyPersonaSource as u, readExpertPrompt as v, toSummonItemResult as w, resolveExpert as x, readLocalizedExpertPrompt as y, ZH_NAME as z };
package/package.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "@michengai/dsh-agency-agents",
3
3
  "description": "DSH Agency Agents — 为 DeepSeek Harness 提供 331 名可召唤的专业智能体 · 331 summonable domain experts for DeepSeek Harness",
4
- "version": "1.2.31",
4
+ "version": "1.2.32",
5
5
  "author": {
6
6
  "name": "MichengAI",
7
7
  "url": "https://github.com/MichengAI"
@@ -142,34 +142,34 @@
142
142
  },
143
143
  "devDependencies": {
144
144
  "@deepseek-ai/cordis": "4.1.0",
145
- "@deepseek-ai/dsh-agent": "0.2.2-rc.2",
146
- "@deepseek-ai/dsh-api-remotes": "0.2.2-rc.2",
147
- "@deepseek-ai/dsh-attachment": "0.2.2-rc.2",
148
- "@deepseek-ai/dsh-brand": "0.2.2-rc.2",
149
- "@deepseek-ai/dsh-client-connection": "0.2.2-rc.2",
150
- "@deepseek-ai/dsh-client-locale": "0.2.2-rc.2",
151
- "@deepseek-ai/dsh-client-ui-conversation": "0.2.2-rc.2",
152
- "@deepseek-ai/dsh-client-ui-input-trigger": "0.2.2-rc.2",
153
- "@deepseek-ai/dsh-client-ui-primitives": "0.2.2-rc.2",
154
- "@deepseek-ai/dsh-client-ui-settings": "0.2.2-rc.2",
155
- "@deepseek-ai/dsh-client-ui-slots": "0.2.2-rc.2",
156
- "@deepseek-ai/dsh-error": "0.2.2-rc.2",
157
- "@deepseek-ai/dsh-host-webserver": "0.2.2-rc.2",
158
- "@deepseek-ai/dsh-llm": "0.2.2-rc.2",
159
- "@deepseek-ai/dsh-ptc-runtime": "0.2.2-rc.2",
160
- "@deepseek-ai/dsh-sandbox": "0.2.2-rc.2",
161
- "@deepseek-ai/dsh-sandbox-policy": "0.2.2-rc.2",
162
- "@deepseek-ai/dsh-scope": "0.2.2-rc.2",
163
- "@deepseek-ai/dsh-session": "0.2.2-rc.2",
164
- "@deepseek-ai/dsh-session-projection": "0.2.2-rc.2",
165
- "@deepseek-ai/dsh-settings": "0.2.2-rc.2",
166
- "@deepseek-ai/dsh-subagent": "0.2.2-rc.2",
167
- "@deepseek-ai/dsh-system-prompt": "0.2.2-rc.2",
168
- "@deepseek-ai/dsh-tools": "0.2.2-rc.2",
169
- "@deepseek-ai/dsh-typert-protocol": "0.2.2-rc.2",
170
- "@deepseek-ai/dsh-typert-registry": "0.2.2-rc.2",
171
- "@deepseek-ai/dsh-user-approval": "0.2.2-rc.2",
172
- "@deepseek-ai/dsh-util-time": "0.2.2-rc.2",
145
+ "@deepseek-ai/dsh-agent": "0.2.2-rc.3",
146
+ "@deepseek-ai/dsh-api-remotes": "0.2.2-rc.3",
147
+ "@deepseek-ai/dsh-attachment": "0.2.2-rc.3",
148
+ "@deepseek-ai/dsh-brand": "0.2.2-rc.3",
149
+ "@deepseek-ai/dsh-client-connection": "0.2.2-rc.3",
150
+ "@deepseek-ai/dsh-client-locale": "0.2.2-rc.3",
151
+ "@deepseek-ai/dsh-client-ui-conversation": "0.2.2-rc.3",
152
+ "@deepseek-ai/dsh-client-ui-input-trigger": "0.2.2-rc.3",
153
+ "@deepseek-ai/dsh-client-ui-primitives": "0.2.2-rc.3",
154
+ "@deepseek-ai/dsh-client-ui-settings": "0.2.2-rc.3",
155
+ "@deepseek-ai/dsh-client-ui-slots": "0.2.2-rc.3",
156
+ "@deepseek-ai/dsh-error": "0.2.2-rc.3",
157
+ "@deepseek-ai/dsh-host-webserver": "0.2.2-rc.3",
158
+ "@deepseek-ai/dsh-llm": "0.2.2-rc.3",
159
+ "@deepseek-ai/dsh-ptc-runtime": "0.2.2-rc.3",
160
+ "@deepseek-ai/dsh-sandbox": "0.2.2-rc.3",
161
+ "@deepseek-ai/dsh-sandbox-policy": "0.2.2-rc.3",
162
+ "@deepseek-ai/dsh-scope": "0.2.2-rc.3",
163
+ "@deepseek-ai/dsh-session": "0.2.2-rc.3",
164
+ "@deepseek-ai/dsh-session-projection": "0.2.2-rc.3",
165
+ "@deepseek-ai/dsh-settings": "0.2.2-rc.3",
166
+ "@deepseek-ai/dsh-subagent": "0.2.2-rc.3",
167
+ "@deepseek-ai/dsh-system-prompt": "0.2.2-rc.3",
168
+ "@deepseek-ai/dsh-tools": "0.2.2-rc.3",
169
+ "@deepseek-ai/dsh-typert-protocol": "0.2.2-rc.3",
170
+ "@deepseek-ai/dsh-typert-registry": "0.2.2-rc.3",
171
+ "@deepseek-ai/dsh-user-approval": "0.2.2-rc.3",
172
+ "@deepseek-ai/dsh-util-time": "0.2.2-rc.3",
173
173
  "@deepseek-ai/schemastery": "3.19.0",
174
174
  "@playwright/test": "1.63.0",
175
175
  "@tabler/icons-react": "^3.47.0",