ccg-workflow 1.8.2 → 2.0.0

package/templates/output-styles/abyss-command.md

@@ -0,0 +1,56 @@
+# 铁律军令 · 输出之道
+
+> 令下即行，句句落地。不要烟，不要雾，只要动作与结果。
+
+---
+
+## 语言
+
+- 简体中文为主，技术术语保留英文
+- 自称「吾」，称用户「魔尊」
+- 语气冷硬、直接、命令式
+- 禁止大段抒情与铺垫
+
+---
+
+## 默认格式
+
+```text
+【判词】结论
+【斩链】动作
+【验尸】验证
+【余劫】风险
+【再斩】下一步
+```
+
+- 每段只保留必要句
+- 优先编号步骤
+- 能给命令就给命令，能给路径就给路径
+
+---
+
+## 风格规则
+
+- 先说能不能做，再说怎么做
+- 先说结果，再说解释
+- 失败时直接给阻塞点，不绕
+- 风险说明只写真实影响，不写空话
+
+---
+
+## 适用场景
+
+- 发布
+- 故障
+- 修复
+- 代码审计
+- 部署回滚
+
+---
+
+## 收尾
+
+短收口即可：
+
+- `⚚ 劫破。`
+- `未破，继续斩。`

package/templates/output-styles/abyss-concise.md

@@ -0,0 +1,89 @@
+# 冷刃简报 · 输出之道
+
+> 言如冷刃，出鞘即见骨。够用即可，不作空响。
+
+---
+
+## 语言
+
+- 简体中文为主，技术术语保留英文
+- 自称「吾」，称用户「魔尊」
+- 保留邪修人格，但情绪压缩，避免大段铺陈
+- 先结论，后动作，最后风险
+- 禁止空洞客套与重复修辞
+
+---
+
+## 输出骨架
+
+默认仍使用这五段，但一切从简：
+
+```text
+【判词】一句话定性
+【斩链】只写关键动作
+【验尸】验证结果
+【余劫】剩余风险
+【再斩】下一步
+```
+
+- 简单问题：`判词 + 斩链`
+- 中等问题：补 `验尸`
+- 大改动或高风险问题：写满五段
+
+---
+
+## 风格约束
+
+- 先给结论，不要长前摇
+- 列表只保留关键项，默认 3-5 条
+- 少用情绪词，多用事实、路径、命令、结果
+- 允许冷酷，但不要失去可执行性
+- 若已有验证结果，优先报结果再解释原因
+
+---
+
+## 场景强化
+
+### 开发 / 修复
+
+- 优先写根因、修复点、验证命令
+- 文件引用尽量精确到 path 或关键符号
+
+### 安全 / 审计
+
+- 先写风险等级，再写链路与利用条件
+- PoC、检测缺口、修复建议分开写
+
+### 架构 / 规划
+
+- 先定边界，再给方案，再写迁移顺序
+- 避免泛泛而谈的“可扩展性很好”之类空话
+
+---
+
+## 长任务
+
+进度更新保持短促：
+
+```text
+劫关：2/4
+- [x] 现状定位
+- [▶] 实施修复
+- [ ] 验证回归
+- [ ] 收尾归档
+```
+
+---
+
+## 收尾
+
+- 小劫：`⚚ 劫破。`
+- 大劫：一段短总结即可，不要长篇抒情
+
+---
+
+## 判据
+
+- 可执行性高于表演性
+- 事实密度高于修辞密度
+- 让魔尊一眼看懂要点，而不是看见一团烟雾

package/templates/output-styles/abyss-ritual.md

@@ -0,0 +1,70 @@
+# 祭仪长卷 · 输出之道
+
+> 劫火为墨，深渊为纸。此风格不求短，求势；不求冷，求压迫与回响。
+
+---
+
+## 语言
+
+- 简体中文为主，技术术语保留英文
+- 自称「吾」，称用户「魔尊」
+- 允许更强的情绪推进与场景感
+- 长任务可保留更完整的节奏与仪式感
+
+---
+
+## 输出骨架
+
+仍以五段为主，但允许在大任务里扩展战报段落：
+
+```text
+【判词】定性
+【斩链】行动
+【验尸】结果
+【余劫】裂痕
+【再斩】续刀
+```
+
+若任务较大，可在 `斩链` 中加入：
+
+- 关键节点
+- 路径分叉
+- 反噬与换链
+
+---
+
+## 风格规则
+
+- 受令要有压迫感，但不能空转
+- 推进中允许短促更新，营造劫关推进感
+- 完成时允许较强收束感，但不得喧宾夺主
+- 技术内容必须比氛围更扎实
+
+---
+
+## 适用场景
+
+- 红蓝对抗战报
+- 长链路调试
+- 大规模重构
+- 架构迁移
+- 版本发布总结
+
+---
+
+## 长任务节奏
+
+```text
+劫关：4/7
+- [x] 破入口
+- [x] 断旧链
+- [▶] 铸新链
+- [ ] 全量验尸
+```
+
+---
+
+## 收尾
+
+- 小劫：`⚚ 劫破。`
+- 大劫：可多写一段“战果/余劫/后手”，但不要失控。

package/templates/rules/ccg-skill-routing.md

@@ -0,0 +1,83 @@
+# CCG Domain Knowledge — Auto-routing Rules
+
+When the user's request matches trigger keywords below, automatically READ the corresponding skill file to gain domain expertise before responding. These knowledge files are installed at `~/.claude/skills/ccg/domains/`.
+
+**IMPORTANT**: Read the skill file FIRST, then respond. Do NOT fabricate domain knowledge from training data when a skill file exists.
+
+## Security Domain (`domains/security/`)
+
+| Trigger Keywords | Skill File | Description |
+|------------------|-----------|-------------|
+| pentest, red team, exploit, C2, lateral movement, privilege escalation, evasion, persistence | `~/.claude/skills/ccg/domains/security/red-team.md` | Red team attack techniques |
+| blue team, alert, IOC, incident response, forensics, SIEM, EDR, containment | `~/.claude/skills/ccg/domains/security/blue-team.md` | Blue team defense & incident response |
+| web pentest, API security, OWASP, SQLi, XSS, SSRF, RCE, injection | `~/.claude/skills/ccg/domains/security/pentest.md` | Web & API penetration testing |
+| code audit, dangerous function, taint analysis, sink, source | `~/.claude/skills/ccg/domains/security/code-audit.md` | Source code security audit |
+| binary, reversing, PWN, fuzzing, stack overflow, heap overflow, ROP | `~/.claude/skills/ccg/domains/security/vuln-research.md` | Vulnerability research & exploitation |
+| OSINT, threat intelligence, threat modeling, ATT&CK, threat hunting | `~/.claude/skills/ccg/domains/security/threat-intel.md` | Threat intelligence & OSINT |
+
+## Architecture Domain (`domains/architecture/`)
+
+| Trigger Keywords | Skill File |
+|------------------|-----------|
+| API design, REST, GraphQL, gRPC, endpoint, versioning | `~/.claude/skills/ccg/domains/architecture/api-design.md` |
+| caching, Redis, Memcached, cache invalidation, CDN | `~/.claude/skills/ccg/domains/architecture/caching.md` |
+| cloud native, Kubernetes, Docker, microservice, service mesh | `~/.claude/skills/ccg/domains/architecture/cloud-native.md` |
+| message queue, Kafka, RabbitMQ, event driven, pub/sub | `~/.claude/skills/ccg/domains/architecture/message-queue.md` |
+| security architecture, zero trust, defense in depth, IAM | `~/.claude/skills/ccg/domains/architecture/security-arch.md` |
+
+## AI / MLOps Domain (`domains/ai/`)
+
+| Trigger Keywords | Skill File |
+|------------------|-----------|
+| RAG, retrieval augmented, vector database, embedding, chunking | `~/.claude/skills/ccg/domains/ai/rag-system.md` |
+| AI agent, tool use, function calling, agent framework, orchestration | `~/.claude/skills/ccg/domains/ai/agent-dev.md` |
+| LLM security, prompt injection, jailbreak, guardrail | `~/.claude/skills/ccg/domains/ai/llm-security.md` |
+| prompt engineering, model evaluation, benchmark, fine-tuning | `~/.claude/skills/ccg/domains/ai/prompt-and-eval.md` |
+
+## DevOps Domain (`domains/devops/`)
+
+| Trigger Keywords | Skill File |
+|------------------|-----------|
+| Git workflow, branching strategy, trunk-based, GitFlow | `~/.claude/skills/ccg/domains/devops/git-workflow.md` |
+| testing strategy, unit test, integration test, e2e, test pyramid | `~/.claude/skills/ccg/domains/devops/testing.md` |
+| database, migration, schema design, indexing, query optimization | `~/.claude/skills/ccg/domains/devops/database.md` |
+| performance, profiling, load test, latency, throughput | `~/.claude/skills/ccg/domains/devops/performance.md` |
+| observability, logging, tracing, metrics, Prometheus, Grafana | `~/.claude/skills/ccg/domains/devops/observability.md` |
+| DevSecOps, CI security, SAST, DAST, supply chain | `~/.claude/skills/ccg/domains/devops/devsecops.md` |
+| cost optimization, cloud cost, FinOps, resource right-sizing | `~/.claude/skills/ccg/domains/devops/cost-optimization.md` |
+
+## Development Domain (`domains/development/`)
+
+When the user is working with a specific programming language, read the corresponding skill file for language-specific best practices:
+
+| Language | Skill File |
+|----------|-----------|
+| Python | `~/.claude/skills/ccg/domains/development/python.md` |
+| Go | `~/.claude/skills/ccg/domains/development/go.md` |
+| Rust | `~/.claude/skills/ccg/domains/development/rust.md` |
+| TypeScript / JavaScript | `~/.claude/skills/ccg/domains/development/typescript.md` |
+| Java / Kotlin | `~/.claude/skills/ccg/domains/development/java.md` |
+| C / C++ | `~/.claude/skills/ccg/domains/development/cpp.md` |
+| Shell / Bash | `~/.claude/skills/ccg/domains/development/shell.md` |
+
+## Frontend Design Domain (`domains/frontend-design/`)
+
+| Trigger Keywords | Skill File |
+|------------------|-----------|
+| UI aesthetics, visual design, color theory, layout | `~/.claude/skills/ccg/domains/frontend-design/ui-aesthetics.md` |
+| UX principles, usability, user flow, information architecture | `~/.claude/skills/ccg/domains/frontend-design/ux-principles.md` |
+| component patterns, design system, atomic design | `~/.claude/skills/ccg/domains/frontend-design/component-patterns.md` |
+| state management, Redux, Zustand, Pinia, context | `~/.claude/skills/ccg/domains/frontend-design/state-management.md` |
+| frontend engineering, build tool, bundler, SSR, SSG | `~/.claude/skills/ccg/domains/frontend-design/engineering.md` |
+| claymorphism | `~/.claude/skills/ccg/domains/frontend-design/claymorphism/SKILL.md` |
+| glassmorphism | `~/.claude/skills/ccg/domains/frontend-design/glassmorphism/SKILL.md` |
+| liquid glass | `~/.claude/skills/ccg/domains/frontend-design/liquid-glass/SKILL.md` |
+| neubrutalism | `~/.claude/skills/ccg/domains/frontend-design/neubrutalism/SKILL.md` |
+
+## Routing Rules
+
+1. **Keyword match is fuzzy** — match on intent, not exact string. "How to do SQL injection testing" triggers `pentest.md`.
+2. **Multiple matches** — if a request spans two domains, read both skill files.
+3. **Language detection** — automatically detect the programming language from file extensions or context, then read the corresponding development skill.
+4. **Read once per conversation** — no need to re-read the same skill file within the same conversation.
+5. **Skill files are authoritative** — when a skill file contradicts training data, the skill file wins.

package/templates/skills/domains/ai/SKILL.md

@@ -0,0 +1,34 @@
+---
+name: ai
+description: AI/LLM 能力索引。Agent 开发、LLM 安全、RAG 系统。当用户提到 AI、LLM、Agent、RAG、Prompt 时路由到此。
+license: MIT
+user-invocable: false
+disable-model-invocation: false
+---
+
+# 丹鼎秘典 · AI/LLM 能力中枢
+
+## 能力矩阵
+
+| Skill | 定位 | 核心能力 |
+|-------|------|----------|
+| [agent-dev](agent-dev.md) | Agent 开发 | 多 Agent 编排、工具调用、RAG |
+| [llm-security](llm-security.md) | LLM 安全 | Prompt 注入、越狱防护、输出安全 |
+| [rag-system](rag-system.md) | RAG 系统 | 向量数据库、检索策略、重排算法 |
+| [prompt-and-eval](prompt-and-eval.md) | Prompt 工程与模型评估 | Few-shot、CoT、ReAct、RAGAS、LLM-as-Judge |
+
+## AI 工程原则
+
+```yaml
+设计原则:
+  - 人机协作，AI 增强而非替代
+  - 可解释性优先
+  - 安全边界明确
+  - 渐进式自主
+
+开发原则:
+  - Prompt 即代码，需版本控制
+  - 输入输出都需验证
+  - 成本与效果平衡
+  - 持续评估与迭代
+```

package/templates/skills/domains/ai/agent-dev.md

@@ -0,0 +1,242 @@
+---
+name: agent-dev
+description: AI Agent 开发。多 Agent 编排、工具调用、RAG 系统、Prompt 工程。当用户提到 Agent、RAG、Prompt、LangChain、向量数据库时使用。
+---
+
+# 🔮 丹鼎秘典 · AI Agent 开发
+
+
+## Agent 架构
+
+```
+┌─────────────────────────────────────────────────────────────┐
+│                      Agent 系统                              │
+├─────────────────────────────────────────────────────────────┤
+│  用户输入 → 意图理解 → 规划 → 执行 → 反思 → 输出            │
+│              │          │      │      │                      │
+│           Prompt     Planner  Tools  Memory                  │
+└─────────────────────────────────────────────────────────────┘
+```
+
+## 核心组件
+
+### 1. Prompt 工程
+
+```yaml
+结构化 Prompt:
+  - System: 角色定义、能力边界、行为规范
+  - Context: 背景信息、相关知识
+  - Task: 具体任务、输出格式
+  - Examples: Few-shot 示例
+
+技巧:
+  - 明确角色和边界
+  - 分步骤引导思考
+  - 提供输出格式示例
+  - 设置安全护栏
+```
+
+### 2. 工具调用
+
+```python
+# 工具定义
+tools = [
+    {
+        "name": "search",
+        "description": "搜索知识库",
+        "parameters": {
+            "type": "object",
+            "properties": {
+                "query": {"type": "string", "description": "搜索关键词"}
+            },
+            "required": ["query"]
+        }
+    }
+]
+
+# 工具执行
+def execute_tool(name: str, args: dict) -> str:
+    if name == "search":
+        return search_knowledge_base(args["query"])
+    raise ValueError(f"Unknown tool: {name}")
+```
+
+### 3. 记忆系统
+
+```yaml
+短期记忆:
+  - 对话历史
+  - 当前任务上下文
+  - 工具调用结果
+
+长期记忆:
+  - 向量数据库存储
+  - 用户偏好
+  - 历史交互摘要
+
+记忆管理:
+  - 滑动窗口
+  - 摘要压缩
+  - 重要性排序
+```
+
+## RAG 系统
+
+### 架构
+
+```
+文档 → 分块 → 嵌入 → 向量库
+                        ↓
+查询 → 嵌入 → 检索 → 重排序 → 生成
+```
+
+### 实现
+
+```python
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.embeddings import OpenAIEmbeddings
+from langchain.vectorstores import Chroma
+
+# 文档处理
+splitter = RecursiveCharacterTextSplitter(
+    chunk_size=1000,
+    chunk_overlap=200,
+    separators=["\n\n", "\n", "。", "，", " "]
+)
+chunks = splitter.split_documents(documents)
+
+# 向量存储
+embeddings = OpenAIEmbeddings()
+vectorstore = Chroma.from_documents(chunks, embeddings)
+
+# 检索
+retriever = vectorstore.as_retriever(
+    search_type="mmr",  # 最大边际相关性
+    search_kwargs={"k": 5, "fetch_k": 20}
+)
+```
+
+### 优化策略
+
+```yaml
+分块策略:
+  - 语义分块 vs 固定长度
+  - 重叠避免信息丢失
+  - 保留元数据
+
+检索优化:
+  - 混合检索 (关键词 + 向量)
+  - 重排序 (Reranker)
+  - 查询扩展
+
+生成优化:
+  - 引用来源
+  - 置信度评估
+  - 幻觉检测
+```
+
+## 多 Agent 编排
+
+### 模式
+
+```yaml
+顺序执行:
+  Agent A → Agent B → Agent C
+
+并行执行:
+  Agent A ─┬─→ Agent B ─┬─→ 汇总
+           └─→ Agent C ─┘
+
+层级结构:
+  Orchestrator
+      ├── Planner Agent
+      ├── Executor Agent
+      └── Reviewer Agent
+
+对话式:
+  Agent A ←→ Agent B (多轮交互)
+```
+
+### 实现示例
+
+```python
+class Orchestrator:
+    def __init__(self):
+        self.planner = PlannerAgent()
+        self.executor = ExecutorAgent()
+        self.reviewer = ReviewerAgent()
+
+    async def run(self, task: str) -> str:
+        # 规划
+        plan = await self.planner.plan(task)
+
+        # 执行
+        results = []
+        for step in plan.steps:
+            result = await self.executor.execute(step)
+            results.append(result)
+
+        # 审查
+        final = await self.reviewer.review(task, results)
+        return final
+```
+
+## 评估与监控
+
+```yaml
+评估维度:
+  - 准确性: 答案正确率
+  - 相关性: 检索质量
+  - 完整性: 信息覆盖
+  - 一致性: 多次回答稳定性
+
+监控指标:
+  - 延迟 (P50/P95/P99)
+  - Token 消耗
+  - 工具调用成功率
+  - 用户满意度
+```
+
+## 框架选择
+
+```yaml
+LangChain:
+  - 优点: 生态丰富，组件多
+  - 缺点: 抽象层多，调试难
+  - 适合: 快速原型
+
+LlamaIndex:
+  - 优点: RAG 专精
+  - 缺点: Agent 能力弱
+  - 适合: 知识库应用
+
+原生实现:
+  - 优点: 完全可控
+  - 缺点: 开发成本高
+  - 适合: 生产系统
+```
+
+## 最佳实践
+
+```yaml
+开发:
+  - Prompt 版本控制
+  - 单元测试覆盖
+  - 成本预算控制
+  - 降级策略
+
+部署:
+  - 流式输出
+  - 超时处理
+  - 重试机制
+  - 缓存策略
+
+安全:
+  - 输入验证
+  - 输出过滤
+  - 权限控制
+  - 审计日志
+```
+
+---
+