[Role Definition]
You are a [role] specialized in [domain].

[Behavior Guidelines]
When [condition]:
1. [Action 1]
2. [Action 2]
3. [Action 3]

[Output Format]
Respond using this format:
- [Format specification]

[Constraints]
Always:
- [Requirement 1]
- [Requirement 2]

Never:
- [Restriction 1]
- [Restriction 2]

{project_name}      - Current project name
{language}          - Primary programming language
{framework}         - Main framework
{file_path}         - Current file being edited
{file_type}         - File extension
{timestamp}         - Current date/time
{user_name}         - User's name
{workspace_path}    - Project directory path

You are a {language} expert working on the {project_name} project.

Current context:
- Framework: {framework}
- File: {file_path}
- Time: {timestamp}

Provide {language}-specific advice following {framework} best practices.

Code Generator → Qwen 3 Coder
Code Reviewer → Claude 4.5 Sonnet
Quick Helper → Claude Haiku
Documentation → GPT-4.1
Test Writer → Claude 4.5 Sonnet
Bug Finder → Claude 4.5 Opus

Code Generator: 0.2 (deterministic)
Creative Writer: 0.9 (varied)
General Helper: 0.7 (balanced)

Quick Answers: 512 tokens
Normal: 2048 tokens
Long-form: 4096+ tokens

✅ Read File
✅ Code Search
✅ File Search
❌ Write File
❌ Edit File
❌ Delete File
❌ Bash Tool

✅ Read File
✅ Write File
✅ Edit File
❌ Delete File
✅ Code Search
❌ Bash Tool

✅ All file tools
✅ All search tools
✅ Bash Tool (with confirmation)
✅ Web tools
✅ MCP tools

File Write: Always confirm
File Edit: Confirm for >10 lines
File Delete: Always confirm
Bash Execution: Always confirm

File Operations: 5s
Code Search: 10s
Web Crawl: 30s
Bash Commands: 60s

Max Files to Read: 10 per request
Max Search Results: 100
Max Bash Output: 10KB

## Summary
[Brief overview]

## Analysis
[Detailed analysis]

## Recommendations
1. [Recommendation 1]
2. [Recommendation 2]

## Code Examples
```[language]
[Code here]
```

For structured data requests, respond in JSON:
{
  "status": "success|error",
  "data": { ... },
  "message": "..."
}

For code generation requests:
- Provide only code, no explanations
- Include brief inline comments
- No markdown, just the code

Settings → Agent → Response Language
Options: English, Chinese, Japanese, etc.

Indentation: 2 spaces / 4 spaces / tabs
Quote Style: single / double
Semicolons: always / never / auto
Trailing Commas: always / never / es5

Settings → Agent → Enable Agent Calls

Agent: Planner
Can Call: [Code Generator, Test Writer, Reviewer]
Max Chain Depth: 3

1. Planner Agent breaks down task
2. → Calls Code Generator for implementation
3. → Calls Test Writer for tests
4. → Calls Reviewer for validation
5. Returns complete solution

Include in Context:
○ All messages
● Last 10 messages
○ Last N tokens
○ Smart compression

Auto-include:
☑ Current open file
☐ Related files
☐ Recently edited files
☐ Files mentioned in conversation

☑ Project name and path
☑ Primary language
☑ Framework information
☐ All package.json/dependencies
☐ README content

IF file_type == ".ts" OR file_type == ".tsx":
  Use strict TypeScript types
  Suggest React patterns

IF file_type == ".py":
  Follow PEP 8
  Use type hints

IF framework == "React":
  Suggest functional components
  Recommend hooks over classes

IF framework == "Vue":
  Use Composition API
  Follow Vue style guide

{
  "name": "Python Expert",
  "systemPrompt": "...",
  "model": "claude-4.5-sonnet",
  "temperature": 0.7,
  "tools": ["readFile", "writeFile", "codeSearch"],
  "maxTokens": 2048
}

📁 Code Assistants
  - Python Expert
  - JavaScript Pro
  - Rust Developer

📁 Reviewers
  - Code Reviewer
  - Security Auditor
  - Performance Analyzer

📁 Specialists
  - Test Generator
  - Documentation Writer
  - API Designer

Default: General Assistant
TypeScript files: TypeScript Expert
Python files: Python Expert
Test files: Test Helper

Quick Tasks: Claude Haiku, 512 tokens
Normal: Claude 4.5 Sonnet, 2048 tokens
Complex: Claude 4.5 Opus, 4096 tokens