feat: add dynamic output format for prompt-enhancer and update README

- Redesign output format to be dynamic and task-adaptive - Replace fixed template with core + optional fields structure - Add simple and complex task examples - Update workflow and enhancement guidelines - Add Agent Skills section to README.md and README_CN.md - Document Prompt Enhancer skill usage with -e/--enhance flags - Provide skill creation best practices 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
2026-02-05 01:50:27 +08:00 · 2025-10-18 23:00:08 +08:00
parent 44734a447c
commit 24aad6238a
3 changed files with 204 additions and 58 deletions
--- a/.claude/skills/prompt-enhancer/SKILL.md
+++ b/.claude/skills/prompt-enhancer/SKILL.md
@@ -18,81 +18,107 @@ allowed-tools: (none)

 ## Output Format

+**Dynamic Structure**: Adapt fields based on task type and context needs. Not all fields are required.
+
+**Core Fields** (always present):
+- **INTENT**: One-sentence technical goal
+- **ACTION**: Concrete steps with technical details
+
+**Optional Fields** (include when relevant):
+- **TECH STACK**: Relevant technologies (when tech-specific)
+- **CONTEXT**: Session memory findings (when context matters)
+- **ATTENTION**: Critical constraints (when risks/requirements exist)
+- **SCOPE**: Affected modules/files (for multi-module tasks)
+- **METRICS**: Success criteria (for optimization/performance tasks)
+- **DEPENDENCIES**: Related components (for integration tasks)
+
+**Example (Simple Task)**:
 ```
 📋 ENHANCED PROMPT

-INTENT: [One-sentence technical goal with tech stack / 明确技术目标含技术栈]
-
-TECH STACK: [Relevant technologies from memory / 相关技术栈]
- [Framework/Library: Purpose]
-
-CONTEXT: [Session memory findings / 会话记忆发现]
- [Key context point 1]
- [Key context point 2]
- [Design patterns/constraints from session]
+INTENT: Fix authentication token validation in JWT middleware

 ACTION:
-1. [Concrete step with technical details / 具体步骤含技术细节]
-2. [Concrete step with technical details]
-3. [Testing/validation step]
+1. Review token expiration logic in auth middleware
+2. Add proper error handling for expired tokens
+3. Test with valid/expired/malformed tokens
+```

-ATTENTION: [Critical constraints / 关键约束]
- [Security/compatibility/performance concerns]
- [Design pattern requirements]
+**Example (Complex Task)**:
+```
+📋 ENHANCED PROMPT
+
+INTENT: Optimize API performance with caching and database indexing
+
+TECH STACK:
+- Redis: Response caching
+- PostgreSQL: Query optimization
+
+CONTEXT:
+- API response times >2s mentioned in previous conversation
+- PostgreSQL slow query logs show N+1 problems
+
+ACTION:
+1. Profile endpoints to identify slow queries
+2. Add PostgreSQL indexes on frequently queried columns
+3. Implement Redis caching for read-heavy endpoints
+4. Add cache invalidation on data updates
+
+METRICS:
+- Target: <500ms API response time
+- Cache hit ratio: >80%
+
+ATTENTION:
+- Maintain backward compatibility with existing API contracts
+- Handle cache invalidation correctly to avoid stale data
 ```
 ## Workflow

 ```
-Trigger → Internal Analysis → Direct Output
-   ↓            ↓                  ↓
-  P1-4    Semantic+Memory    Enhanced Prompt
-          (3 dimensions)      (Structured)
+Trigger (-e/--enhance) → Internal Analysis → Dynamic Output
+         ↓                       ↓                  ↓
+   User Input           Assess Task Type      Select Fields
+                    Extract Memory Context    Structure Prompt
 ```

-1. **Detect**: Check triggers (P1-P4)
-2. **Internal Analysis**:
-   - Semantic (EN/CN) intent analysis
-   - Memory extraction (tech stack, patterns, constraints)
-   - Enhancement (structure + supplement + clarify)
-3. **Output**: Present enhanced structured prompt directly
+1. **Detect**: User input contains `-e` or `--enhance`
+2. **Analyze**:
+   - Determine task type (fix/optimize/implement/refactor)
+   - Extract relevant session context
+   - Identify tech stack and constraints
+3. **Structure**:
+   - Always include: INTENT + ACTION
+   - Conditionally add: TECH STACK, CONTEXT, ATTENTION, METRICS, etc.
+4. **Output**: Present dynamically structured prompt

-## Enhancement Checklist (Internal)
+## Enhancement Guidelines (Internal)

-**Structure**:
- [ ] INTENT: Clear, one-sentence technical goal
- [ ] TECH STACK: Relevant technologies from session
- [ ] CONTEXT: Key session findings and constraints
- [ ] ACTION: Concrete steps with technical details
- [ ] ATTENTION: Critical constraints and patterns
+**Always Include**:
+- Clear, actionable INTENT
+- Concrete ACTION steps with technical details

-**Supplement**:
- [ ] Add tech stack/frameworks mentioned in session
- [ ] Include design patterns if relevant
- [ ] Add testing/validation requirements
- [ ] Specify performance metrics if applicable
+**Add When Relevant**:
+- TECH STACK: Task involves specific technologies
+- CONTEXT: Session memory provides useful background
+- ATTENTION: Security/compatibility/performance concerns exist
+- SCOPE: Multi-module or cross-component changes
+- METRICS: Performance/optimization goals need measurement
+- DEPENDENCIES: Integration points matter

-**Clarify**:
- [ ] Make vague intent explicit
- [ ] Resolve ambiguous references (it/that/这个/那个)
- [ ] Expand multi-module scope with dependencies
- [ ] Add missing context from memory
+**Quality Checks**:
+- Make vague intent explicit
+- Resolve ambiguous references
+- Add testing/validation steps
+- Include constraints from memory

 ## Best Practices

- ✅ Detect `-e`/`--enhance` flags first (P1)
- ✅ Support EN + CN semantic keywords
+- ✅ Trigger only on `-e`/`--enhance` flags
+- ✅ Use **dynamic field selection** based on task type
 - ✅ Extract **memory context ONLY** (no file reading)
- ✅ Add tech stack, design patterns, testing requirements
- ✅ Direct output (no intermediate steps)
- ✅ Use INTENT/TECH STACK/CONTEXT/ACTION/ATTENTION format
- ❌ NO tool calls (AskUserQuestion removed)
- ❌ NO Bash, Read, Glob, Grep operations
- ❌ NO file analysis or codebase scanning
-
-## Key Changes
-
-1. **Removed all tools** - Pure analysis and output
-2. **Removed user confirmation** - Direct output for speed
-3. **Added tech stack section** - Supplement with technologies
-4. **Enhanced internal analysis** - 3 dimensions (structure + supplement + clarify)
-5. **Focus on memory** - Session context only, no file reading
+- ✅ Always include INTENT + ACTION as core fields
+- ✅ Add optional fields only when relevant to task
+- ✅ Direct output (no intermediate steps shown)
+- ❌ NO tool calls
+- ❌ NO file operations (Bash, Read, Glob, Grep)
+- ❌ NO fixed template - adapt to task needs
--- a/README.md
+++ b/README.md
@@ -289,6 +289,66 @@ MCP (Model Context Protocol) tools provide advanced codebase analysis. **Recomme

 ---

+## 🎯 Agent Skills
+
+**Agent Skills** extend Claude's capabilities with modular, reusable expertise packages. Skills are automatically discovered and triggered based on context.
+
+### **Available Skills**
+
+#### **Prompt Enhancer**
+Transform vague prompts into structured, actionable specifications.
+
+**Usage**: Add `-e` or `--enhance` flag to any request
+```bash
+# Example
+fix authentication issue -e
+optimize database queries --enhance
+```
+
+**What it does**:
+- Analyzes session context and tech stack
+- Structures intent into INTENT → ACTION format
+- Adds relevant context, constraints, and metrics
+- Outputs ready-to-execute specifications
+
+**Output**: Dynamic structured prompt adapted to task type (simple tasks get minimal fields, complex tasks get comprehensive specifications)
+
+### **Creating Custom Skills**
+
+Skills are stored in `.claude/skills/` (project) or `~/.claude/skills/` (personal):
+
+```bash
+# Create a new skill
+mkdir -p .claude/skills/my-skill
+```
+
+**SKILL.md structure**:
+```yaml
+---
+name: My Skill Name
+description: What it does and when to use it. Use when [trigger conditions].
+allowed-tools: Read, Grep, Glob  # Optional tool restrictions
+---
+
+# My Skill Name
+
+## Instructions
+[Step-by-step guidance for Claude]
+
+## Examples
+[Concrete usage examples]
+```
+
+**Best Practices**:
+- Write specific descriptions with clear trigger conditions
+- Include both what the skill does and when to use it
+- Keep skills focused on single capabilities
+- Test by asking questions that match your description
+
+For complete documentation, see [skills.md](skills.md).
+
+---
+
 ## 🚀 Getting Started

 ### Complete Development Workflow
--- a/README_CN.md
+++ b/README_CN.md
@@ -289,6 +289,66 @@ MCP (模型上下文协议) 工具提供高级代码库分析。**推荐安装**

 ---

+## 🎯 智能体技能 (Agent Skills)
+
+**智能体技能** 通过模块化、可复用的专业知识包扩展 Claude 的能力。技能会根据上下文自动发现和触发。
+
+### **可用技能**
+
+#### **提示词增强器 (Prompt Enhancer)**
+将模糊提示转换为结构化、可执行的规范。
+
+**使用方式**：在任何请求中添加 `-e` 或 `--enhance` 标志
+```bash
+# 示例
+修复认证问题 -e
+优化数据库查询 --enhance
+```
+
+**功能说明**:
+- 分析会话上下文和技术栈
+- 将意图结构化为 INTENT → ACTION 格式
+- 添加相关上下文、约束和指标
+- 输出可立即执行的规范
+
+**输出**：根据任务类型动态调整的结构化提示（简单任务使用最少字段，复杂任务使用全面规范）
+
+### **创建自定义技能**
+
+技能存储在 `.claude/skills/`（项目）或 `~/.claude/skills/`（个人）：
+
+```bash
+# 创建新技能
+mkdir -p .claude/skills/my-skill
+```
+
+**SKILL.md 结构**:
+```yaml
+---
+name: 我的技能名称
+description: 功能说明和使用时机。使用时机：[触发条件]。
+allowed-tools: Read, Grep, Glob  # 可选工具限制
+---
+
+# 我的技能名称
+
+## 指令
+[Claude 的分步指导]
+
+## 示例
+[具体使用示例]
+```
+
+**最佳实践**:
+- 编写具体的描述，包含明确的触发条件
+- 同时说明功能和使用时机
+- 保持技能专注于单一能力
+- 通过匹配描述的问题进行测试
+
+完整文档请参阅 [skills.md](skills.md)。
+
+---
+
 ## 🚀 快速入门

 ### 完整开发工作流