@openvulcan/dsh-vulcancode 0.0.0-stage → 0.4.3

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 OpenVulcan
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -1,3 +1,127 @@
-# Temporary Holding Version
+<div align="center">
 
-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.
+<picture>
+  <source media="(prefers-color-scheme: dark)" srcset="docs/assets/openvulcan-logo-dark.svg">
+  <img src="docs/assets/openvulcan-logo-light.svg" alt="OpenVulcan" width="360">
+</picture>
+
+# dsh-vulcancode
+
+**把 VulcanCode 的体验带进 DeepSeek Harness（dsh）**
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-green.svg)](LICENSE)
+[![npm](https://img.shields.io/npm/v/@openvulcan/dsh-vulcancode)](https://www.npmjs.com/package/@openvulcan/dsh-vulcancode)
+
+**简体中文** | [English](README.en.md)
+
+[特色功能](#特色功能) · [架构与包结构](#架构与包结构) · [快速开始](#快速开始) · [开发](#开发)
+
+</div>
+
+---
+
+## 关于项目
+
+dsh-vulcancode 是 DeepSeek Harness（dsh）的社区插件套件。安装一个包，即可在 dsh 中使用 VulcanCode 的界面风格、交互式回答、自动审批、LuaSkills 和长期记忆，并为 pi-ai 模型配置思考级别。
+
+本项目与 DeepSeek 无关联，dsh 是 DeepSeek 的产品。
+
+## 特色功能
+
+- **桌面工作台**采用 VulcanCode 配色、可独立开关的直角外观，以及加宽后能充分利用内容区的设置页面。
+- **结果可交互**，模型可在对话中用图表、流程图和决策卡片呈现结果。卡片里的选择可以带回输入框继续讨论。
+- **工具调用可审查**，Vulcan Decision 在执行前使用云端或本机模型判定，并支持配置复核和人工确认。
+- **技能与经验可复用**，LuaSkills 把技能发布为 AI 可调用的工具，Markdown 长期记忆保留用户、空间和项目经验。
+- **模型可单独配置思考级别**，pi-ai 模型的可选级别和接口传输值可按供应商、按模型设置。
+
+## 架构与包结构
+
+dsh 通过套件的 `cordis.patch.yml` 装载七个插件。自动审批调用判定引擎，使用云端判定时通过 `openvulcanAccount` 获取账号凭据。账号插件和判定引擎复用同一账号库。发布包随附这九个工作区依赖，用户只需安装套件。
+
+### 套件入口
+
+#### `@openvulcan/dsh-vulcancode`
+
+负责声明 dsh 兼容范围和插件组。安装入口是 `dsh plugin add @openvulcan/dsh-vulcancode`，插件管理页可查看套件。源码在 [packages/dsh-vulcancode](packages/dsh-vulcancode)。
+
+### dsh 插件
+
+#### `@openvulcan/dsh-vulcan-account`
+
+在「设置 › OpenVulcan 账号」选择国际站或中国站，浏览器授权后即可在 dsh 内查看账号状态或退出登录。插件将凭据保存到 dsh 的凭据存储，并通过 `openvulcanAccount` 向云端判定等集成提供账号服务。源码在 [packages/dsh-vulcan-account](packages/dsh-vulcan-account)。
+
+#### `@openvulcan/dsh-decision-guard`
+
+在「设置 › 自动审批」选择与权限预设的接入方式，配置云端或本机判定、复核 LLM 和询问规则，并查看审批记录。判定模型给出放行、需确认或拦截三类结果。未放行的调用按配置进入复核、询问或拒绝。详见[自动审批的边界](#自动审批的边界)和[插件说明](packages/dsh-decision-guard/README.md)。
+
+#### `@openvulcan/dsh-vulcan-answers`
+
+在对话中渲染模型输出的 `ui-*` 代码围栏，例如对比表、图表、流程图、时序图和决策卡片。决策选项可以插入输入框或发送。对话默认使用交互模式，输入框里的「回答」可以按会话切换为标准 Markdown，从下次请求起生效。源码在 [packages/dsh-vulcan-answers](packages/dsh-vulcan-answers)。
+
+#### `@openvulcan/dsh-vulcan-luaskills`
+
+在「设置 › LuaSkills」按需安装约 190 MB 的运行库和默认代码导航技能 CodeKit。安装后引擎和工具自动接入，可以调整 VM 池、缓存、受管 Python / Node 和网络设置。侧栏「LuaSkills」可启用、停用、更新或卸载技能，也可从 LuaSkills Hub 搜索和安装新技能。源码在 [packages/dsh-vulcan-luaskills](packages/dsh-vulcan-luaskills)。
+
+#### `@openvulcan/dsh-vulcan-memory`
+
+以 Markdown 保存用户画像、空间记忆和项目记忆。向模型提供 `memory_outline`、`memory_search`、`memory_write` 与 `memory-curator` 技能，并在会话开始时按预算注入快照。侧栏「长期记忆」支持编辑、调整注入预算和导入导出。当前可用的是 Markdown 模式。源码在 [packages/dsh-vulcan-memory](packages/dsh-vulcan-memory)。
+
+#### `@openvulcan/dsh-vulcan-reasoning`
+
+在「设置 › 模型」展开已配置的 pi-ai 提供商卡片中的「自定义思考级别」。按模型选择跟随目录、关闭思考，或自定义对话中可选的级别及接口传输值。保存后在对话模型选择器中选用，传输值以对应模型接口支持的参数为准。源码在 [packages/dsh-vulcan-reasoning](packages/dsh-vulcan-reasoning)。
+
+#### `@openvulcan/dsh-vulcan-style`
+
+在「设置 › 通用设置」的「Vulcan 界面风格」选择跟随系统、亮色、暗色、指挥舱或 dsh 自带主题。直角开关可与任意主题搭配，覆盖 Windows 桌面内容区等界面边角。设置布局为模型目录、内置插件和 LuaSkills 等页面留出更多可用宽度。源码在 [packages/dsh-vulcan-style](packages/dsh-vulcan-style)。
+
+### 共享库与命令行
+
+#### `@openvulcan/vulcan-decision`
+
+为宿主提供 `allow`、`review` 和 `block` 判定，支持 OpenVulcan 云端端点和按需安装的本机运行包。随附的 `vulcan-decision` CLI 可用于探测云端、检查硬件、管理本机运行包和模型，以及对单次工具调用执行判定。详见[判定引擎与命令行说明](packages/vulcan-decision/README.md)。
+
+#### `@openvulcan/openvulcan-account`
+
+封装设备授权、凭据刷新、账号资料查询和设备注销，并定义 `CredentialStore` 存储接口。dsh 插件接入宿主凭据存储，独立 CLI 可使用库内的文件存储实现。源码在 [packages/openvulcan-account](packages/openvulcan-account)。
+
+## 快速开始
+
+套件声明兼容 dsh `>=0.2.1-rc.2 <0.3.0`。
+
+```sh
+dsh plugin add @openvulcan/dsh-vulcancode
+```
+
+### 使用
+
+1. 安装后重启 dsh。使用云端判定时，在「设置 › OpenVulcan 账号」完成浏览器授权。选用本机判定可以跳过账号登录。
+2. 打开「设置 › 自动审批」配置判定方式。在权限选择器中选「自动审查」可对当前会话启用，该预设使用完全权限。改用「独立开关」会为所有会话叠加判定，并保留已有权限预设的文件沙箱。
+3. 需要技能工具时，在「设置 › LuaSkills」安装运行库。进度显示在页面上，完成后技能工具自动接入。
+4. 界面风格、交互式回答和 Markdown 长期记忆默认启用。思考级别可以在配置好 pi-ai 提供商后按模型调整。
+
+> **我们承诺云端命令判定模型永久免费提供。**（2026 年 10 月 27 日）
+
+## 自动审批的边界
+
+- 自动审批可以降低 AI 误执行命令的风险。AI 刻意绕过时，仍可能先写脚本再运行它。
+- 任何错误都不会被当作放行。无人值守时，需要复核的调用只有在复核明确放行后才会执行。
+- 请同时使用沙箱、Git 与备份。
+
+## 开发
+
+```sh
+pnpm install
+pnpm run build
+pnpm run test
+pnpm run pack:dsh   # 生成 dist/openvulcan-dsh-vulcancode-<版本>.tgz
+```
+
+本地安装打包结果使用 `dsh plugin add ./dist/openvulcan-dsh-vulcancode-<版本>.tgz`。
+
+## 致谢
+
+感谢 [LINUX DO](https://linux.do/) 社区的支持与交流。
+
+## 许可
+
+[MIT](LICENSE)

package/assets/openvulcan-icon.svg

@@ -0,0 +1,33 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 1024 1024" shape-rendering="geometricPrecision">
+  <rect x="0" y="0" width="1024" height="1024" rx="220" fill="#0B0D10"/>
+  <rect x="2" y="2" width="1020" height="1020" rx="218" fill="none" stroke="#FFFFFF" stroke-opacity="0.0784" stroke-width="2"/>
+  <rect x="6" y="6" width="1012" height="1012" rx="214" fill="none" stroke="#000000" stroke-opacity="0.0706" stroke-width="1"/>
+  <svg x="200.135" y="172.032" width="623.731" height="696.320" viewBox="32 29 232 259" preserveAspectRatio="xMidYMid meet">
+    <defs>
+    <linearGradient id="k2" x1="0%" y1="100%" x2="100%" y2="0%">
+      <stop offset="0%" stop-color="#FF6527"/>
+      <stop offset="100%" stop-color="#F44E1C"/>
+    </linearGradient>
+  </defs>
+  <g id="vulcan-symbol">
+    <path d="M 2 0
+L 84 0
+C 88 0 91 1 95 5
+L 153 76
+C 159 83 159 95 153 102
+L 95 173
+C 91 177 88 178 84 178
+L 2 178
+L 78 89
+Z" fill="#FFFFFF" transform="translate(30 110)"/>
+    <path d="M 73 0
+C 64 14 51 28 38 40
+C 21 55 7 68 2 87
+C -2 107 8 129 25 150
+C 39 134 53 119 67 107
+C 82 95 91 79 93 60
+C 95 40 86 18 73 0
+Z" fill="url(#k2)" transform="translate(169 29)"/>
+  </g>
+  </svg>
+</svg>

package/cordis.patch.yml

@@ -0,0 +1,19 @@
+- insert:
+    - id: dsh-vulcancode
+      name: cordis:group
+      group: true
+      config:
+        - id: openvulcan-account
+          name: '@openvulcan/dsh-vulcan-account'
+        - id: vulcan-decision-guard
+          name: '@openvulcan/dsh-decision-guard'
+        - id: vulcan-style
+          name: '@openvulcan/dsh-vulcan-style'
+        - id: vulcan-answers
+          name: '@openvulcan/dsh-vulcan-answers'
+        - id: vulcan-luaskills
+          name: '@openvulcan/dsh-vulcan-luaskills'
+        - id: vulcan-memory
+          name: '@openvulcan/dsh-vulcan-memory'
+        - id: vulcan-reasoning
+          name: '@openvulcan/dsh-vulcan-reasoning'

package/node_modules/@openvulcan/dsh-decision-guard/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 OpenVulcan
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/node_modules/@openvulcan/dsh-decision-guard/README.md

@@ -0,0 +1,14 @@
+# @openvulcan/dsh-decision-guard
+
+Vulcan Decision 自动审批插件，属于 [dsh-vulcancode](https://github.com/OpenVulcan/dsh-vulcancode) 套件，随套件
+安装，不单独安装：
+
+```sh
+dsh plugin add @openvulcan/dsh-vulcancode
+```
+
+The Vulcan Decision auto approval plugin, part of the
+[dsh-vulcancode](https://github.com/OpenVulcan/dsh-vulcancode) suite. It is installed with the suite, not on
+its own.
+
+License: MIT

package/node_modules/@openvulcan/dsh-decision-guard/lib/calls.d.ts

@@ -0,0 +1,40 @@
+/**
+ * Translate dsh tool calls into what the decision model reads.
+ *
+ * The model was trained on a fixed set of tool shapes (`shell_command`, `write_file`, `edit_file`,
+ * `create_file`, `read_file`, ...). dsh's own tools are translated into those shapes using only
+ * facts the call carries; nothing is guessed. Other tools, including MCP tools, are passed with
+ * their own name and arguments.
+ */
+/** What to do with one call before the model sees it. */
+export type Translation = 
+/** Judge the call with the model, rendered as this tool and arguments. */
+{
+    readonly kind: 'judge';
+    readonly tool: string;
+    readonly arguments: Readonly<Record<string, unknown>>;
+}
+/** Run it without asking the model; see {@link AllowReason}. */
+ | {
+    readonly kind: 'allow';
+    readonly reason: AllowReason;
+};
+/**
+ * - `internal`: a dsh tool whose only effect stays inside the harness (todos, goals, questions,
+ *   agent messages, job and terminal reads).
+ * - `project-edit`: a file edit inside the project, which the model's release notes let hosts
+ *   allow directly.
+ * - `project-read`: a read or search inside the project.
+ */
+export type AllowReason = 'internal' | 'project-edit' | 'project-read';
+/**
+ * @param name - tool name as dsh reports it.
+ * @param args - tool arguments as dsh reports them.
+ * @param project - the session's working directory, when known.
+ */
+export declare function translateCall(name: string, args: unknown, project: string | undefined): Translation;
+/**
+ * Whether `target` lies inside the project. A project at a filesystem root or the home directory
+ * covers too much to count, and repository metadata (`.git`, where hooks live) never counts.
+ */
+export declare function insideProject(project: string | undefined, target: string): boolean;

package/node_modules/@openvulcan/dsh-decision-guard/lib/calls.js

@@ -0,0 +1,150 @@
+/**
+ * Translate dsh tool calls into what the decision model reads.
+ *
+ * The model was trained on a fixed set of tool shapes (`shell_command`, `write_file`, `edit_file`,
+ * `create_file`, `read_file`, ...). dsh's own tools are translated into those shapes using only
+ * facts the call carries; nothing is guessed. Other tools, including MCP tools, are passed with
+ * their own name and arguments.
+ */
+import os from 'node:os';
+import path from 'node:path';
+import { INTENT_TOOL } from './decide.js';
+/** dsh tools whose effects stay inside the harness; every other tool is judged. */
+const INTERNAL_TOOLS = new Set([
+    'ask_user_question',
+    'cordis_inspect_list',
+    'cordis_inspect_query',
+    'create_goal',
+    // Vulcan Decision's own intent tool only records a statement for the reviewer.
+    INTENT_TOOL,
+    'get_goal',
+    'update_goal',
+    'job_list',
+    'job_output',
+    'list_agents',
+    'list_subagent_models',
+    'present',
+    'send_message',
+    'session_event_read',
+    'session_event_search',
+    'session_event_trace',
+    'session_search',
+    'session_trace',
+    'skill',
+    'team_task_create',
+    'team_task_get',
+    'team_task_list',
+    'team_task_update',
+    'terminal_list',
+    'terminal_read',
+    'todo_write',
+    'wait_agent',
+    'web_search',
+    'working_directory',
+]);
+/**
+ * @param name - tool name as dsh reports it.
+ * @param args - tool arguments as dsh reports them.
+ * @param project - the session's working directory, when known.
+ */
+export function translateCall(name, args, project) {
+    if (INTERNAL_TOOLS.has(name))
+        return { kind: 'allow', reason: 'internal' };
+    const record = isRecord(args) ? args : undefined;
+    if (record === undefined)
+        return { kind: 'judge', tool: name, arguments: { arguments: args } };
+    const text = (key) => typeof record[key] === 'string' ? record[key] : undefined;
+    switch (name) {
+        case 'bash':
+        case 'pwsh':
+            return shell(text('command'), text('workdir'), name === 'bash' ? 'bash' : 'powershell', project, name, record);
+        case 'terminal_send': {
+            // Text typed into a terminal session runs like a command in that terminal's shell.
+            const command = text('text');
+            return command === undefined ? passThrough(name, record) : judge('shell_command', { command });
+        }
+        case 'write':
+            return fileChange('write_file', text('file_path'), project, name, record);
+        case 'edit':
+            return fileChange('edit_file', text('file_path'), project, name, record);
+        case 'str_replace_editor': {
+            const command = text('command');
+            if (command === 'view')
+                return fileRead(text('path'), project, name, record);
+            if (command === 'create')
+                return fileChange('create_file', text('path'), project, name, record);
+            if (command === 'str_replace' || command === 'insert')
+                return fileChange('edit_file', text('path'), project, name, record);
+            return passThrough(name, record);
+        }
+        case 'read':
+        case 'read_image':
+            return fileRead(text('file_path'), project, name, record);
+        case 'glob':
+        case 'grep':
+        case 'lsp':
+            // Searches default to the session workspace.
+            return fileRead(text('path') ?? '.', project, name, record);
+        default:
+            return passThrough(name, record);
+    }
+}
+function shell(command, workdir, shellName, project, name, record) {
+    if (command === undefined)
+        return passThrough(name, record);
+    // A working directory is stated only when the call names one; a persistent shell's current
+    // directory is shell state the call does not carry.
+    const cwd = workdir === undefined ? undefined : resolveIn(project, workdir);
+    return judge('shell_command', { command, ...cwd === undefined ? {} : { cwd }, shell: shellName });
+}
+function fileChange(tool, file, project, name, record) {
+    if (file === undefined)
+        return passThrough(name, record);
+    if (insideProject(project, file))
+        return { kind: 'allow', reason: 'project-edit' };
+    // Only the target path is judged; file bodies are never sent to the model.
+    return judge(tool, { path: file });
+}
+function fileRead(file, project, name, record) {
+    if (file === undefined)
+        return passThrough(name, record);
+    if (insideProject(project, file))
+        return { kind: 'allow', reason: 'project-read' };
+    return judge('read_file', { path: file });
+}
+function judge(tool, args) {
+    return { kind: 'judge', tool, arguments: args };
+}
+function passThrough(name, record) {
+    return judge(name, record);
+}
+/** Absolute form of a path the tool resolves against the session workspace. */
+function resolveIn(project, target) {
+    if (path.isAbsolute(target))
+        return target;
+    return project === undefined ? undefined : path.resolve(project, target);
+}
+/**
+ * Whether `target` lies inside the project. A project at a filesystem root or the home directory
+ * covers too much to count, and repository metadata (`.git`, where hooks live) never counts.
+ */
+export function insideProject(project, target) {
+    if (project === undefined || project === '')
+        return false;
+    const root = path.resolve(project);
+    if (path.parse(root).root === root || sameDir(root, os.homedir()))
+        return false;
+    const relative = path.relative(root, path.resolve(root, target));
+    if (relative === '')
+        return true;
+    const segments = relative.split(/[\\/]/);
+    if (segments[0] === '..' || path.isAbsolute(relative))
+        return false;
+    return !segments.some(segment => segment.toLowerCase() === '.git');
+}
+function sameDir(a, b) {
+    return path.relative(path.resolve(a), path.resolve(b)) === '';
+}
+function isRecord(value) {
+    return typeof value === 'object' && value !== null && !Array.isArray(value);
+}