securityclaw 0.0.1

package/CHANGELOG.md

@@ -0,0 +1,49 @@
+# Changelog
+
+## [0.1.0] - 2026-03-15
+
+### 架构重构
+
+#### 新增
+- 引入分层架构（领域层、应用层、基础设施层）
+- 新增 `ContextInferenceService` - 上下文推断服务（350 行）
+- 新增 `ApprovalService` - 审批业务逻辑服务（200 行）
+- 新增 `NotificationAdapter` - 7 个消息渠道适配器（150 行）
+- 新增 `ApprovalCommands` - 命令处理器（180 行）
+- 新增 `OpenClawAdapter` - OpenClaw API 封装（60 行）
+- 新增 `FormattingService` - 格式化工具（60 行）
+- 新增 `ApprovalSubjectResolver` - 审批主题解析（35 行）
+- 新增 `PluginConfigParser` - 配置解析（130 行）
+
+#### 接口定义
+- 新增 `NotificationPort` - 通知接口
+- 新增 `ApprovalRepository` - 审批仓储接口
+- 新增 `OpenClawAdapter` - OpenClaw 适配器接口
+
+#### 改进
+- 代码行数减少 35.7%（2032 → 1305 行）
+- 最大文件行数减少 82.8%（2032 → 350 行）
+- 实现单一职责原则
+- 实现依赖倒置原则
+- 提升可测试性和可维护性
+
+#### 修复
+- 修复 `escapeRegExp` 函数错误（`src/engine/rule_engine.ts:23`）
+- 修复正则表达式未闭合问题（`src/engine/rule_engine.ts:46`）
+
+#### 文档
+- 新增 `README.md` - 项目说明
+- 新增 `docs/ARCHITECTURE.md` - 架构文档
+- 更新 `docs/TECHNICAL_SOLUTION.md` - 包含重构后架构
+- 更新 `docs/RUNBOOK.md` - 包含新组件使用示例
+
+### 测试
+- ✅ 所有测试通过（48/48）
+- ✅ 类型检查通过（0 errors）
+- ✅ 向后兼容（100%）
+- ✅ 无性能损失
+
+### 性能
+- Hook 延迟 p95: ~5ms
+- 内存占用: ~20MB
+- 启动时间: ~50ms

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 SecurityClaw contributors
+
+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

@@ -0,0 +1,135 @@
+# SecurityClaw Security Plugin
+
+[中文文档](./README.zh-CN.md)
+
+SecurityClaw is a runtime security plugin for [OpenClaw](https://github.com/openclaw/openclaw). It enforces policy decisions on tool calls, supports approval workflows, sanitizes sensitive outputs, and exposes audit-ready decision telemetry.
+
+## Why SecurityClaw
+
+LLM agents can execute powerful tools. SecurityClaw provides a policy guardrail layer so risky operations are either blocked, challenged for approval, or allowed with warning and traceability.
+
+## Core Capabilities
+
+- Runtime policy enforcement for OpenClaw hooks (`before_tool_call`, `after_tool_call`, etc.)
+- Rule-first security model (`allow`, `warn`, `challenge`, `block`)
+- Challenge approval workflow with command-based admin handling
+- Dynamic sensitive-path registry that maps paths to asset labels before rule evaluation
+- Sensitive data scanning and sanitization (DLP)
+- Admin dashboard for strategy and account policy operations
+- Decision events for audit and observability
+- Built-in internationalization (`en` and `zh-CN`) for runtime/admin text
+
+## Architecture
+
+SecurityClaw follows a layered architecture:
+
+- `domain`: policy, approval, context inference, formatting
+- `domain/services/sensitive_path_registry.ts`: built-in + runtime-overridden sensitive path mappings
+- `engine`: rule matching, decisioning, DLP scanning
+- `config`: base YAML + SQLite runtime override
+- `admin`: dashboard backend + frontend
+- `monitoring`: runtime status and decision snapshots
+
+See [Architecture](./docs/ARCHITECTURE.md) and [Technical Solution](./docs/TECHNICAL_SOLUTION.md).
+
+## Quick Start
+
+### 1. Install dependencies
+
+```bash
+npm install
+```
+
+### 2. Install into OpenClaw
+
+```bash
+npm run openclaw:install
+```
+
+Alternative install paths for end users:
+
+```bash
+npx securityclaw install
+curl -fsSL https://raw.githubusercontent.com/znary/securityclaw/main/install.sh | bash
+```
+
+### 3. Run verification
+
+```bash
+npm test
+```
+
+### 4. Start admin dashboard (standalone)
+
+```bash
+npm run admin
+```
+
+Default dashboard URL: `http://127.0.0.1:4780`
+
+## OpenClaw Integration
+
+Preferred local install:
+
+```bash
+npm run openclaw:install
+```
+
+This creates a versioned plugin archive, installs it through `openclaw plugins install`, restarts the gateway, and verifies gateway health.
+See [OpenClaw Install Guide](./docs/OPENCLAW_INSTALL.md) for details.
+
+## Approval Commands
+
+After setting one account policy with `is_admin=true`, the admin can run:
+
+- `/securityclaw-approve <approval_id>`
+- `/securityclaw-approve <approval_id> long`
+- `/securityclaw-reject <approval_id>`
+- `/securityclaw-pending`
+
+## Admin Dashboard
+
+Dashboard supports English and Chinese UI switching and stores language preference in local storage.
+By default, it follows the host system language.
+
+Main panels:
+
+- Overview: posture and trend signals, plus a skill-risk snapshot for high-priority installed skills
+- Decisions: recent decision events and reasons
+- Policies: grouped rule strategy controls plus sensitive-path registry management
+- Skill Interception: installed skill inventory, risk scoring, undeclared-change detection, rescan/quarantine/trust override actions, and interception policy matrix
+- Accounts: admin approver account selection and mode settings
+
+Sensitive path registry behavior:
+
+- Built-in path patterns cover credentials, personal content, download staging, browser profiles, browser secret stores, and communication stores.
+- Registry entries are persisted in SQLite runtime strategy overrides together with rule decisions.
+- Built-in entries can be disabled from the dashboard, and custom path rules can be added without editing the base YAML.
+
+Skill interception behavior:
+
+- Dashboard discovers installed skills from local OpenClaw / Codex skill roots and stores scan results in SQLite.
+- A skill can be flagged when its content changes without a matching version update.
+- Overview surfaces the most important skill signals directly so admins can see high-risk items without switching tabs.
+- The dedicated Skill Interception panel supports rescan, quarantine, temporary trust override, and risk-matrix editing.
+
+## Documentation
+
+- [Documentation Index](./docs/README.md)
+- [OpenClaw Install Guide](./docs/OPENCLAW_INSTALL.md)
+- [Admin Dashboard](./docs/ADMIN_DASHBOARD.md)
+- [Runbook](./docs/RUNBOOK.md)
+- [Integration Guide](./docs/INTEGRATION_GUIDE.md)
+
+## Development
+
+```bash
+npm run typecheck
+npm run test:unit
+npm test
+npm run admin:build
+```
+
+## License
+
+MIT. See [LICENSE](./LICENSE).

package/README.zh-CN.md

@@ -0,0 +1,135 @@
+# SecurityClaw 安全插件
+
+[English](./README.md)
+
+SecurityClaw 是面向 [OpenClaw](https://github.com/openclaw/openclaw) 的运行时安全插件。它在工具调用链路上执行安全策略，支持审批流程、敏感信息净化与审计级决策记录。
+
+## SecurityClaw 解决什么问题
+
+LLM Agent 具备高权限工具调用能力。SecurityClaw 在运行时提供策略护栏，将高风险操作按规则执行为拦截、审批确认、提醒或放行，并保留可追溯审计信息。
+
+## 核心能力
+
+- 基于 OpenClaw Hook 的运行时策略执行（`before_tool_call` 等）
+- 规则优先决策模型（`allow` / `warn` / `challenge` / `block`）
+- Challenge 审批流程与管理员命令处理
+- 动态敏感路径注册表，在规则判断前先把路径映射成资产标签
+- DLP 扫描与敏感输出净化
+- 管理后台（策略与账号策略配置）
+- 决策事件与状态观测
+- 中英文国际化（`en` / `zh-CN`）
+
+## 架构说明
+
+分层结构如下：
+
+- `domain`：策略、审批、上下文推断、格式化
+- `domain/services/sensitive_path_registry.ts`：内置 + 运行时覆写的敏感路径映射
+- `engine`：规则匹配、决策引擎、DLP
+- `config`：YAML 基线配置 + SQLite 运行时覆盖
+- `admin`：管理后台前后端
+- `monitoring`：运行状态与决策快照
+
+详见 [架构文档](./docs/ARCHITECTURE.md) 与 [技术方案](./docs/TECHNICAL_SOLUTION.md)。
+
+## 快速开始
+
+### 1. 安装依赖
+
+```bash
+npm install
+```
+
+### 2. 安装到 OpenClaw
+
+```bash
+npm run openclaw:install
+```
+
+终端用户也可以直接使用：
+
+```bash
+npx securityclaw install
+curl -fsSL https://raw.githubusercontent.com/znary/securityclaw/main/install.sh | bash
+```
+
+### 3. 执行验证
+
+```bash
+npm test
+```
+
+### 4. 启动管理后台（独立模式）
+
+```bash
+npm run admin
+```
+
+默认地址：`http://127.0.0.1:4780`
+
+## OpenClaw 集成
+
+推荐本地安装方式：
+
+```bash
+npm run openclaw:install
+```
+
+这条命令会生成带版本号的插件压缩包，通过 `openclaw plugins install` 安装，随后自动重启 gateway 并校验状态。
+详情见 [OpenClaw 安装指南](./docs/OPENCLAW_INSTALL.md)。
+
+## 审批命令
+
+当账号策略中配置 `is_admin=true` 后，管理员可在聊天渠道执行：
+
+- `/securityclaw-approve <approval_id>`
+- `/securityclaw-approve <approval_id> long`
+- `/securityclaw-reject <approval_id>`
+- `/securityclaw-pending`
+
+## 管理后台
+
+管理后台支持中英文切换，并将语言偏好保存在本地存储。
+默认跟随系统语言。
+
+核心模块：
+
+- 概览：总体态势、趋势，以及高优先级已安装 skill 的风险快照
+- 决策记录：最近决策事件与原因
+- 规则策略：按分组编辑规则动作，并维护敏感路径注册表
+- Skill 拦截：已安装 skill 清单、风险打分、未声明变更检测、重扫 / 隔离 / 受信覆盖操作，以及拦截策略矩阵
+- 账号策略：管理员审批账号与模式配置
+
+敏感路径注册表说明：
+
+- 内置覆盖凭据目录、个人内容目录、下载暂存区、浏览器资料目录、浏览器密钥库和通信存储。
+- 路径注册表与规则动作一起持久化到 SQLite 运行时策略覆盖中。
+- 可在后台删除内置项，也可直接添加自定义路径，无需手改基线 YAML。
+
+Skill 拦截说明：
+
+- 后台会从本地 OpenClaw / Codex skill 目录自动发现已安装 skills，并把扫描结果持久化到 SQLite。
+- 当 skill 内容发生变化、但版本号没有同步更新时，会被标记为“内容变了但版本没变”。
+- 概览页会直接展示最值得优先处理的 skill 风险信号，不需要先切到 Skill 页签。
+- `Skill 拦截` 面板支持重扫、隔离、临时受信覆盖，以及风险矩阵配置。
+
+## 文档导航
+
+- [文档索引](./docs/README.zh-CN.md)
+- [OpenClaw 安装指南](./docs/OPENCLAW_INSTALL.md)
+- [管理后台说明](./docs/ADMIN_DASHBOARD.md)
+- [运行手册](./docs/RUNBOOK.md)
+- [集成指南](./docs/INTEGRATION_GUIDE.md)
+
+## 开发命令
+
+```bash
+npm run typecheck
+npm run test:unit
+npm test
+npm run admin:build
+```
+
+## 许可证
+
+MIT，详见 [LICENSE](./LICENSE)。