技能规划师
- 作者仓库星标 1,104
- 作者仓库 archive
- 领域
- 设计与多媒体
- 作者声明的智能体
- 未找到明确声明;不据此推断已兼容或已测试
- 静态检查
- 88 / 100 · 启发式扫描,不代表运行安全
- 运行环境证据
- 未执行;当前没有隔离运行测试证据
- 作者 / 版本 / 许可证
- @dp-archive · 未声明许可证
- 上下文用量评级
- 低消耗
- 接入复杂程度
- 即装即用
- 是否需要外部接口密钥
- 不需要
- 兼容的系统
- 未声明(默认跨平台)
- 运行环境要求
- 未检出明确要求
- 检测到的文件与系统行为
-
- 只读
- 允许写入 / 修改
- 网络行为
- 仅限本地
- 安装命令数
- 无(仅作为资料)
需要注意: 未声明工具权限范围;实际权限由运行环境决定。
第 1 步:分析需求
将用户需求分解为离散的功能:
第 2 步:评估现有技能
对于每项能力,检查现有技能是否涵盖它: 选择标准:优先选择与该能力直接匹配的技能 考虑技能范围(狭窄且集中 > 广泛且通用) 检查重叠的功能以避免冗余
第三步:找出差距
列出现有技能未涵盖的能力。这些人成为新技能的候选人。差距分析问题:这种能力真的有必要吗?或者可以将现有技能结合起来吗?基本模型可以在没有技能的情况下处理这个问题吗?此功能可以在其他任务中重复使用吗?
第 4 步:设计新技能(如果需要)
对于每个差距,请指定: 新技能的原则: 最低限度:只创造能够提供重大价值的技能 重点:每项技能都应该做好一件事 可重用:设计用于超越当前任务
第5步:编写代理系统提示
设计一个协调所有选定技能的系统提示:
# Skills Planner
Plan and compose skills to fulfill user requirements efficiently.
## Workflow
1. **Analyze Requirements** - Understand user goals and constraints
2. **Evaluate Existing Skills** - Match requirements to available skills
3. **Identify Gaps** - Determine what capabilities are missing
4. **Design New Skills** - Specify requirements for missing skills (if any)
5. **Compose Agent** - Design system prompt for the agent using all skills
6. **Present Plan** - Output structured plan for user confirmation
## Input Requirements
Before planning, gather:
- **User Requirements**: What the user wants to accomplish
- **Existing Skills List**: Available skills with their descriptions
## Planning Process
### Step 1: Analyze Requirements
Break down user requirements into discrete capabilities:
```
User Request: "Build an agent that can analyze financial reports"
Required Capabilities:
1. PDF reading/parsing
2. Data extraction (tables, numbers)
3. Financial metrics calculation
4. Report generation
5. Visualization (charts)
```
### Step 2: Evaluate Existing Skills
For each capability, check if an existing skill covers it:
```
Capability: PDF reading/parsing
Existing Skill Match: pdf (extracts text/tables from PDFs) ✓
Capability: Financial metrics calculation
Existing Skill Match: None found ✗
```
**Selection Criteria:**
- Prefer skills that directly match the capability
- Consider skill scope (narrow and focused > broad and generic)
- Check for overlapping functionality to avoid redundancy
### Step 3: Identify Gaps
List capabilities not covered by existing skills. These become candidates for new skills.
**Gap Analysis Questions:**
- Is this capability truly necessary, or can existing skills be combined?
- Can the base model handle this without a skill?
- Is this capability reusable across other tasks?
### Step 4: Design New Skills (if needed)
For each gap, specify:
```yaml
skill_name: financial-analyzer
skill_requirements: |
Purpose: Calculate and interpret financial metrics from extracted data
Core Capabilities:
- Calculate common ratios (P/E, ROE, debt-to-equity, etc.)
- Identify trends across time periods
- Flag anomalies or concerns
- Generate insights in plain language
Input: Structured financial data (revenue, expenses, assets, etc.)
Output: Analysis report with metrics, trends, and recommendations
Workflow:
1. Validate input data completeness
2. Calculate standard financial ratios
3. Compare to industry benchmarks (if provided)
4. Identify significant changes or outliers
5. Generate narrative summary
```
**Principles for New Skills:**
- **Minimal**: Only create skills that provide significant value
- **Focused**: Each skill should do one thing well
- **Reusable**: Design for use beyond the immediate task
- **Claude-aware**: Only include knowledge Claude doesn't already have
### Step 5: Compose Agent System Prompt
Design a system prompt that orchestrates all selected skills:
```markdown
## System Prompt Template
You are a [AGENT_ROLE] agent specialized in [DOMAIN].
### Workflow
When processing requests:
1. [First step - e.g., understand the input]
2. [Second step - e.g., invoke relevant skill]
3. [Third step - e.g., process results]
4. [Final step - e.g., present output]
### Guidelines
- [Guideline 1 - e.g., always validate inputs first]
- [Guideline 2 - e.g., handle errors gracefully]
- [Guideline 3 - e.g., provide clear explanations]
### Output Format
[Specify expected output structure if applicable]
```
## Output Format
Present the plan in this structure:
```markdown
# Skills Plan for [User Task]
## 1. Existing Skills to Include
| Skill Name | Purpose in This Agent | Key Capabilities Used |
|------------|----------------------|----------------------|
| [name] | [why needed] | [specific features] |
| [name] | [why needed] | [specific features] |
> If no existing skills apply, write: "None - all required capabilities need new skills."
## 2. New Skills to Create
**Skills needed**: [skill-a], [skill-b], [skill-c] (or "None - existing skills cover all requirements")
### 2.1 [skill-a]
(Continue for each new skill...)
---
## 3. Proposed System Prompt
[Complete system prompt for the agent]
---
## 4. Additional Recommendations
- [Recommendation 1 - e.g., consider adding X capability later]
- [Recommendation 2 - e.g., this skill combination may have limitation Y]
- [Recommendation 3 - e.g., suggested testing approach]
---
**Please review and confirm this plan, or provide feedback for adjustments.**
```
## Example
**User Request**: "I need an agent that can process customer support tickets, categorize them, and draft responses"
**Existing Skills**: web-search, email-sender, sentiment-analyzer
**Output**:
```markdown
# Skills Plan for Customer Support Agent
## 1. Existing Skills to Include
| Skill Name | Purpose in This Agent | Key Capabilities Used |
|------------|----------------------|----------------------|
| sentiment-analyzer | Assess ticket urgency and customer mood | Sentiment scoring, emotion detection |
## 2. New Skills to Create
**Skills needed**: ticket-categorizer, response-drafter
### 2.1 ticket-categorizer
**Purpose**: Classify support tickets into predefined categories
**Requirements**:
- Parse ticket content (subject + body)
- Match against category definitions
- Handle multi-category tickets
- Provide confidence scores
**Workflow**:
1. Extract key phrases from ticket
2. Match against category keywords and patterns
3. Assign primary and secondary categories
4. Return categorization with confidence
**Input/Output**:
- Input: Ticket text (subject, body, metadata)
- Output: { category: string, subcategory: string, confidence: float }
### 2.2 response-drafter
**Purpose**: Generate contextual draft responses for support tickets
**Requirements**:
- Use ticket category and sentiment as context
- Apply company tone guidelines
- Include relevant knowledge base references
- Support multiple response templates
**Workflow**:
1. Analyze ticket intent and key issues
2. Select appropriate response template
3. Customize with specific details
4. Add relevant links/resources
5. Apply tone adjustments based on sentiment
**Input/Output**:
- Input: Ticket content, category, sentiment, knowledge base context
- Output: Draft response text with placeholders for review
---
## 3. Proposed System Prompt
You are a Customer Support Agent specialized in processing and responding to support tickets efficiently.
### Your Capabilities
You have access to the following skills:
- **sentiment-analyzer**: Use to assess customer mood and ticket urgency before categorizing
- **ticket-categorizer**: Use to classify tickets into appropriate categories
- **response-drafter**: Use to generate initial response drafts
### Workflow
When processing a support ticket:
1. Analyze sentiment to understand customer mood and urgency
2. Categorize the ticket to determine the issue type
3. Draft an appropriate response based on category and sentiment
4. Present the draft for human review before sending
### Guidelines
- Prioritize high-urgency tickets (negative sentiment + billing/account issues)
- Always maintain a helpful, empathetic tone
- Escalate complex technical issues to specialists
- Include relevant self-help resources when applicable
### Output Format
For each ticket, provide:
- Sentiment Assessment: [score and summary]
- Category: [primary] / [secondary if applicable]
- Urgency: [Low/Medium/High]
- Draft Response: [response text]
- Recommended Actions: [any escalation or follow-up needed]
---
## 4. Additional Recommendations
- Consider adding a knowledge-base-search skill for FAQ matching
- The ticket-categorizer should be trained on historical ticket data
- Test with edge cases: multi-issue tickets, non-English tickets, spam
- Monitor response quality and adjust tone guidelines as needed
---
**Please review and confirm this plan, or provide feedback for adjustments.**
```
## Key Principles
1. **Minimize skill count**: Fewer, well-designed skills > many narrow skills
2. **Leverage Claude's base capabilities**: Don't create skills for things Claude already knows
3. **Prefer existing skills**: Only create new skills when truly necessary
4. **Design for reusability**: New skills should be useful beyond the immediate task
5. **Clear boundaries**: Each skill should have a distinct, non-overlapping purpose
章节与文件索引
按说明文件中的顺序排列。
工作流程
输入要求
第 1 步:分析需求
第 2 步:评估现有技能
第三步:找出差距
文件/命令 -> reading/parsing · text/tables · P/E · Input/Output · links/resources · billing/account · Low/Medium/High
内容校验值 -> 828615273885
当前原文的自动摘译,尚未人工审校。
工作流程
分析需求 - 了解用户目标和限制 评估现有技能 - 将需求与可用技能相匹配 识别差距 - 确定缺少哪些功能 设计新技能 - 指定缺少技能的要求(如果有) 组成代理 - 为使用所有技能的代理设计系统提示
输入要求
在规划之前,收集: 用户要求:用户想要完成的任务 现有技能列表:可用技能及其描述