AgentSkillsCN

startproject

以研究与规划为起点,开启新功能的开发进程。通过协调 Claude 的研究子智能体与 Codex 的专业咨询,推动项目的顺利开展。触发条件包括:“/startproject”“启动项目”“新功能”。

SKILL.md
--- frontmatter
name: startproject
description: |
  Initialize new feature development with research and planning.
  Orchestrates Claude research subagents and Codex consultation.
  Triggers: "/startproject", "start project", "new feature".
metadata:
  short-description: Feature initialization workflow

Start Project Skill - Feature Initialization

Orchestrate the start of new feature development.

Workflow

Phase 1: Research (Claude Subagents)

  1. Codebase Exploration

    code
    Task(subagent_type="Explore", prompt="""
    Explore codebase for:
    - Related existing implementations
    - Patterns used in similar features
    - Dependencies and interfaces
    
    Return: File paths and key patterns found
    """)
    
  2. External Research (if needed)

    • WebSearch for library documentation
    • WebFetch for specific API references

Phase 2: Requirements Gathering

  1. Ask user clarifying questions
  2. Document requirements
  3. Identify constraints

Phase 3: Design Consultation (Codex)

code
Task(subagent_type="general-purpose", prompt="""
Consult Codex for design:

codex exec --model o3 --sandbox read-only --full-auto "
Design approach for: {feature}

Research findings:
{research summary}

Requirements:
{requirements}

Recommend:
1. Architecture approach
2. Key design decisions
3. Potential issues
4. Implementation order
"

Return concise design summary.
""")

Phase 4: Task Creation

  1. Create task list based on plan
  2. Set task dependencies
  3. Present for user approval

Output Structure

markdown
## Project: {feature name}

### Research Summary
- Existing related code: {files}
- Patterns identified: {patterns}
- External resources: {links}

### Requirements
- {requirement 1}
- {requirement 2}

### Design (Codex Review)
{Codex recommendation summary}

### Implementation Tasks
1. [ ] {task 1}
2. [ ] {task 2}
3. [ ] {task 3}

### Next Steps
{What to do after approval}

Example Usage

User: "/startproject Add MLflow experiment comparison feature"

  1. Explore existing MLflow integration code
  2. Research MLflow comparison APIs
  3. Ask about specific comparison requirements
  4. Consult Codex for design
  5. Create task breakdown
  6. Present for approval