Pipeline Prompt System Guide

View Source

Table of Contents

  1. Overview
  2. Prompt Types Reference
  3. File-Based Prompt Management
  4. Prompt Template Standards
  5. Component Library Standards
  6. Advanced Prompt Patterns
  7. Content Processing Features
  8. Best Practices
  9. Complete Examples
  10. Migration Guide

Overview

The Pipeline Prompt System provides a comprehensive framework for managing, reusing, and composing prompts across AI workflows. This system enables:

  • External Prompt Files: Store prompts in dedicated files for reusability
  • Dynamic Composition: Combine multiple prompt sources into complex instructions
  • Template Libraries: Standardized prompt components for common patterns
  • Content Processing: Advanced extraction and transformation of prompt content
  • Version Control: Track and manage prompt evolution over time

Architecture

pipeline_ex/
├── pipelines/
│   ├── prompts/           # Reusable prompt templates
│   │   ├── analysis/      # Analysis-focused prompts
│   │   ├── generation/    # Content generation prompts
│   │   ├── extraction/    # Data extraction prompts
│   │   └── validation/    # Quality check prompts
│   └── components/        # Reusable step components
│       ├── validation_steps.yaml
│       ├── transformation_steps.yaml
│       └── llm_steps.yaml
└── workflows/             # Complete workflow definitions
    ├── development/       # Development workflows
    ├── analysis/          # Analysis workflows
    └── production/        # Production workflows

Prompt Types Reference

1. Static Content (static)

Inline text content defined directly in the YAML:

prompt:
  - type: "static"
    content: |
      Analyze this code for the following criteria:
      1. Security vulnerabilities
      2. Performance issues
      3. Code maintainability
      4. Best practice adherence

Use Cases:

  • Short, workflow-specific instructions
  • Connecting text between other prompt types
  • Conditional logic instructions

2. File Content (file)

Load content from external files:

prompt:
  - type: "file"
    path: "pipelines/prompts/analysis/security_review.md"
  - type: "file"
    path: "src/main.py"

Features:

  • Automatic file change detection and caching
  • Support for any text-based file format
  • Relative and absolute path support
  • Error handling for missing files

Use Cases:

  • Reusable prompt templates
  • Loading source code for analysis
  • Loading documentation or requirements
  • Standard prompt components

3. Previous Response (previous_response)

Reference outputs from earlier workflow steps:

prompt:
  - type: "previous_response"
    step: "code_analysis"
  - type: "previous_response"
    step: "security_scan"
    extract: "vulnerabilities"

Fields:

  • step (required): Name of the previous step
  • extract (optional): Extract specific JSON field
  • extract_with (optional): Use ContentExtractor for processing
  • summary (optional): Generate summary of content
  • max_length (optional): Limit content length

Use Cases:

  • Building on previous analysis
  • Passing structured data between steps
  • Context accumulation across workflow

4. Session Context (session_context)

Reference conversation history from Claude sessions:

prompt:
  - type: "session_context"
    session_id: "code_review_session"
    include_last_n: 5

Fields:

  • session_id (required): Session identifier
  • include_last_n (optional): Number of recent messages to include

Use Cases:

  • Multi-turn conversations
  • Maintaining context across session restarts
  • Referencing earlier decisions in long workflows

5. Claude Continue (claude_continue)

Continue existing Claude conversations with new prompts:

prompt:
  - type: "claude_continue"
    new_prompt: "Now add comprehensive error handling to the implementation"

Use Cases:

  • Extending existing implementations
  • Iterative development workflows
  • Progressive enhancement patterns

File-Based Prompt Management

Directory Structure Standards

/pipelines/prompts/ - Prompt Template Library

Organized by purpose and domain:

pipelines/prompts/
├── analysis/
│   ├── code_review.md
│   ├── security_audit.md
│   ├── performance_analysis.md
│   └── dependency_check.md
├── generation/
│   ├── api_documentation.md
│   ├── test_generation.md
│   ├── code_scaffolding.md
│   └── tutorial_creation.md
├── extraction/
│   ├── data_parsing.md
│   ├── entity_extraction.md
│   └── content_summarization.md
├── validation/
│   ├── quality_checks.md
│   ├── compliance_review.md
│   └── output_validation.md
└── common/
    ├── system_prompts.md
    ├── error_handling.md
    └── context_setup.md

Prompt File Naming Conventions

  • Descriptive names: security_vulnerability_scan.md not scan.md
  • Action-oriented: generate_api_tests.md not api_tests.md
  • Domain prefixes: frontend_component_analysis.md, backend_service_review.md
  • Version suffixes: code_review_v2.md for major updates

Prompt File Structure

Each prompt file should follow this template:

# Prompt Title

## Purpose
Brief description of what this prompt accomplishes.

## Context Requirements
- List of required context or input data
- Expected format of inputs
- Prerequisites or dependencies

## Variables
Document any template variables used:
- `{PROJECT_TYPE}` - Type of project being analyzed
- `{LANGUAGE}` - Programming language
- `{FRAMEWORK}` - Framework or library being used

## Prompt Content

[Main prompt content here, using clear sections and examples]

## Expected Output Format
Description of expected response structure.

## Usage Examples
Reference to workflows that use this prompt.

## Version History
- v1.0 (2024-07-03): Initial version
- v1.1 (2024-07-04): Added error handling instructions

Template Variables System

Support for dynamic prompt content:

# Security Analysis Prompt

Analyze the {LANGUAGE} {PROJECT_TYPE} for security vulnerabilities.

Focus areas for {FRAMEWORK} projects:
- Authentication mechanisms
- Data validation
- Error handling
- Dependency security

## Code to Analyze
{CODE_CONTENT}

## Previous Findings
{PREVIOUS_ANALYSIS}

Usage in workflows:

prompt:
  - type: "file"
    path: "pipelines/prompts/analysis/security_review.md"
    variables:
      LANGUAGE: "Python"
      PROJECT_TYPE: "web application"
      FRAMEWORK: "FastAPI"
  - type: "file"
    path: "src/main.py"
    inject_as: "CODE_CONTENT"
  - type: "previous_response"
    step: "initial_scan"
    inject_as: "PREVIOUS_ANALYSIS"

Prompt Template Standards

Template Categories

1. Analysis Templates (/analysis/)

Code Review Template (code_review.md):

# Code Review Analysis

## Objective
Perform comprehensive code review focusing on quality, security, and maintainability.

## Review Criteria
1. **Code Quality**
   - Readability and clarity
   - Proper naming conventions
   - Code organization and structure
   - Documentation completeness

2. **Security Assessment**
   - Input validation
   - Authentication and authorization
   - Data handling and storage
   - Dependency vulnerabilities

3. **Performance Considerations**
   - Algorithm efficiency
   - Resource usage
   - Scalability factors
   - Caching strategies

4. **Maintainability**
   - Code modularity
   - Test coverage
   - Error handling
   - Configuration management

## Output Format
Provide analysis in JSON format:

{ "overall_score": 85, "categories": {

"quality": {"score": 90, "issues": []},
"security": {"score": 80, "issues": []},
"performance": {"score": 85, "issues": []},
"maintainability": {"score": 85, "issues": []}

}, "critical_issues": [], "recommendations": [], "next_steps": [] }

Security Audit Template (security_audit.md):

# Security Vulnerability Assessment

## Scope
Comprehensive security analysis covering OWASP Top 10 and industry best practices.

## Assessment Areas
1. **Authentication & Authorization**
   - User authentication mechanisms
   - Session management
   - Access control implementation
   - Multi-factor authentication

2. **Data Protection**
   - Data encryption at rest and in transit
   - Sensitive data handling
   - PII protection measures
   - Database security

3. **Input Validation**
   - SQL injection prevention
   - XSS protection
   - CSRF safeguards
   - Input sanitization

4. **Infrastructure Security**
   - Server configuration
   - Network security
   - Dependency management
   - Container security

## Risk Classification
- **Critical**: Immediate security risk, requires urgent attention
- **High**: Significant vulnerability, should be addressed soon
- **Medium**: Potential security concern, plan for resolution
- **Low**: Minor security improvement opportunity

## Output Requirements
Security assessment report with:
- Executive summary
- Detailed findings by category
- Risk-prioritized recommendations
- Remediation timeline suggestions

2. Generation Templates (/generation/)

API Documentation Template (api_documentation.md):

# API Documentation Generator

## Objective
Generate comprehensive API documentation from source code and specifications.

## Documentation Requirements
1. **API Overview**
   - Purpose and scope
   - Authentication methods
   - Base URLs and versioning
   - Rate limiting information

2. **Endpoint Documentation**
   - HTTP methods and paths
   - Request/response schemas
   - Parameter descriptions
   - Example requests and responses
   - Error codes and messages

3. **Data Models**
   - Schema definitions
   - Validation rules
   - Relationship mappings
   - Example payloads

4. **Integration Guides**
   - Getting started tutorial
   - Common use cases
   - Code examples in multiple languages
   - Troubleshooting guide

## Output Format
Generate documentation in OpenAPI 3.0 specification format with accompanying Markdown guides.

Test Generation Template (test_generation.md):

# Comprehensive Test Suite Generator

## Testing Strategy
Generate tests covering unit, integration, and end-to-end scenarios.

## Test Categories
1. **Unit Tests**
   - Function-level testing
   - Edge case coverage
   - Error condition handling
   - Mock and stub usage

2. **Integration Tests**
   - API endpoint testing
   - Database integration
   - External service mocking
   - Configuration testing

3. **End-to-End Tests**
   - User journey testing
   - Cross-browser compatibility
   - Performance benchmarks
   - Security testing

## Code Coverage Requirements
- Target: 90%+ line coverage
- 100% coverage for critical paths
- Include negative test cases
- Test error handling paths

## Test Framework Selection
Choose appropriate testing frameworks based on technology stack and project requirements.

3. Extraction Templates (/extraction/)

Data Parsing Template (data_parsing.md):

# Structured Data Extraction

## Extraction Objectives
Parse and structure unstructured or semi-structured data into standardized formats.

## Supported Input Formats
- Plain text documents
- CSV and TSV files
- JSON and XML data
- Log files and reports
- Configuration files

## Extraction Patterns
1. **Entity Recognition**
   - Names, dates, and identifiers
   - Technical specifications
   - Configuration parameters
   - Error messages and codes

2. **Relationship Mapping**
   - Dependency relationships
   - Hierarchical structures
   - Sequential processes
   - Cross-references

3. **Data Validation**
   - Format compliance
   - Completeness checks
   - Consistency validation
   - Quality scoring

## Output Schema
Provide extracted data in JSON format with metadata about extraction confidence and validation results.

4. Validation Templates (/validation/)

Quality Assurance Template (quality_checks.md):

# Quality Assurance Validation

## Quality Metrics
Comprehensive evaluation of deliverable quality across multiple dimensions.

## Evaluation Criteria
1. **Functional Correctness**
   - Requirements compliance
   - Feature completeness
   - Business logic accuracy
   - User experience quality

2. **Technical Excellence**
   - Code quality standards
   - Architecture compliance
   - Performance benchmarks
   - Security requirements

3. **Documentation Quality**
   - Completeness and accuracy
   - Clarity and organization
   - Example quality
   - Maintenance procedures

4. **Process Compliance**
   - Development standards
   - Review procedures
   - Testing requirements
   - Deployment readiness

## Validation Output
Quality scorecard with pass/fail status, improvement recommendations, and certification readiness assessment.

Component Library Standards

/pipelines/components/ - Reusable Step Components

Component Categories

Validation Steps (validation_steps.yaml):

# Reusable validation step components
components:
  code_quality_check:
    type: "gemini"
    role: "brain"
    model: "gemini-2.5-flash"
    token_budget:
      max_output_tokens: 2048
      temperature: 0.3
    prompt:
      - type: "file"
        path: "pipelines/prompts/validation/quality_checks.md"
      - type: "previous_response"
        step: "${source_step}"
    functions:
      - "evaluate_quality"
    output_to_file: "quality_assessment.json"

  security_validation:
    type: "gemini"
    role: "brain"
    model: "gemini-2.5-flash"
    token_budget:
      max_output_tokens: 4096
      temperature: 0.2
    prompt:
      - type: "file"
        path: "pipelines/prompts/validation/security_audit.md"
      - type: "file"
        path: "${code_path}"
    functions:
      - "assess_security"
    output_to_file: "security_report.json"

  compliance_check:
    type: "gemini"
    role: "brain"
    prompt:
      - type: "file"
        path: "pipelines/prompts/validation/compliance_review.md"
      - type: "previous_response"
        step: "${implementation_step}"
        extract: "deliverables"
    output_to_file: "compliance_status.json"

Transformation Steps (transformation_steps.yaml):

components:
  data_normalizer:
    type: "gemini"
    role: "brain"
    model: "gemini-2.5-flash"
    prompt:
      - type: "file"
        path: "pipelines/prompts/extraction/data_parsing.md"
      - type: "file"
        path: "${input_data_path}"
    functions:
      - "normalize_data"
    output_to_file: "normalized_data.json"

  content_summarizer:
    type: "gemini"
    role: "brain"
    token_budget:
      max_output_tokens: 1024
      temperature: 0.4
    prompt:
      - type: "file"
        path: "pipelines/prompts/extraction/content_summarization.md"
      - type: "previous_response"
        step: "${source_step}"
        summary: true
        max_length: 2000
    output_to_file: "content_summary.json"

  format_converter:
    type: "claude"
    role: "muscle"
    claude_options:
      max_turns: 5
      allowed_tools: ["Write", "Read"]
      output_format: "json"
    prompt:
      - type: "static"
        content: "Convert the following data to ${target_format} format:"
      - type: "previous_response"
        step: "${source_step}"
    output_to_file: "converted_${target_format}.${target_extension}"

LLM Steps (llm_steps.yaml):

components:
  smart_analysis:
    type: "claude_smart"
    preset: "analysis"
    claude_options:
      max_turns: 3
      allowed_tools: ["Read"]
    prompt:
      - type: "file"
        path: "pipelines/prompts/analysis/${analysis_type}.md"
      - type: "file"
        path: "${target_file}"
    output_to_file: "${analysis_type}_result.json"

  robust_implementation:
    type: "claude_robust"
    retry_config:
      max_retries: 3
      backoff_strategy: "exponential"
      fallback_action: "simplified_prompt"
    claude_options:
      max_turns: 20
      allowed_tools: ["Write", "Edit", "Read", "Bash"]
      output_format: "json"
    prompt:
      - type: "file"
        path: "pipelines/prompts/generation/${implementation_type}.md"
      - type: "previous_response"
        step: "${planning_step}"
    output_to_file: "${implementation_type}_result.json"

  session_continuation:
    type: "claude_session"
    session_config:
      persist: true
      session_name: "${workflow_name}_session"
      max_turns: 50
    prompt:
      - type: "claude_continue"
        new_prompt: "${continuation_instruction}"
    output_to_file: "session_result.json"

Component Usage Patterns

Including Components in Workflows:

workflow:
  name: "comprehensive_analysis"
  
  steps:
    - name: "initial_scan"
      type: "gemini"
      prompt:
        - type: "file"
          path: "src/main.py"

    # Use validation component
    - <<: *code_quality_check
      name: "quality_assessment"
      variables:
        source_step: "initial_scan"

    # Use transformation component  
    - <<: *content_summarizer
      name: "summary_generation"
      variables:
        source_step: "quality_assessment"

    # Use LLM component with customization
    - <<: *smart_analysis
      name: "detailed_analysis"
      variables:
        analysis_type: "security_audit"
        target_file: "src/main.py"
      claude_options:
        max_turns: 5  # Override component default

Advanced Prompt Patterns

1. Progressive Enhancement Pattern

Build complexity gradually through connected prompts:

steps:
  - name: "basic_analysis"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/analysis/basic_code_review.md"
      - type: "file"
        path: "src/main.py"

  - name: "detailed_analysis"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/analysis/detailed_security_audit.md"
      - type: "previous_response"
        step: "basic_analysis"
        extract: "concerns"
      - type: "file"
        path: "src/main.py"

  - name: "comprehensive_report"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/generation/comprehensive_report.md"
      - type: "previous_response"
        step: "basic_analysis"
      - type: "previous_response"
        step: "detailed_analysis"

2. Context Accumulation Pattern

Build rich context across multiple steps:

steps:
  - name: "requirements_analysis"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/analysis/requirements_review.md"
      - type: "file"
        path: "requirements.md"

  - name: "architecture_review"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/analysis/architecture_analysis.md"
      - type: "file"
        path: "architecture.md"
      - type: "previous_response"
        step: "requirements_analysis"
        extract: "constraints"

  - name: "implementation_plan"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/generation/implementation_planning.md"
      - type: "static"
        content: "Requirements Analysis:"
      - type: "previous_response"
        step: "requirements_analysis"
      - type: "static"
        content: "\nArchitecture Review:"
      - type: "previous_response"
        step: "architecture_review"

3. Iterative Refinement Pattern

Refine outputs through multiple iterations:

steps:
  - name: "initial_draft"
    type: "claude"
    claude_options:
      max_turns: 10
      allowed_tools: ["Write"]
    prompt:
      - type: "file"
        path: "pipelines/prompts/generation/initial_implementation.md"
      - type: "file"
        path: "requirements.md"

  - name: "review_draft"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/validation/implementation_review.md"
      - type: "previous_response"
        step: "initial_draft"

  - name: "refine_implementation"
    type: "claude_session"
    session_config:
      persist: true
      continue_on_restart: true
    prompt:
      - type: "claude_continue"
        new_prompt: |
          Based on this review feedback, please refine the implementation:
      - type: "previous_response"
        step: "review_draft"
        extract: "improvement_suggestions"

4. Parallel Processing Pattern

Process multiple aspects simultaneously:

steps:
  - name: "parallel_analysis"
    type: "parallel_claude"
    parallel_tasks:
      - id: "security_analysis"
        claude_options:
          max_turns: 15
          allowed_tools: ["Read"]
        prompt:
          - type: "file"
            path: "pipelines/prompts/analysis/security_focus.md"
          - type: "file"
            path: "src/main.py"
        output_to_file: "security_analysis.json"

      - id: "performance_analysis"
        claude_options:
          max_turns: 15
          allowed_tools: ["Read"]
        prompt:
          - type: "file"
            path: "pipelines/prompts/analysis/performance_focus.md"
          - type: "file"
            path: "src/main.py"
        output_to_file: "performance_analysis.json"

      - id: "maintainability_analysis"
        claude_options:
          max_turns: 15
          allowed_tools: ["Read"]
        prompt:
          - type: "file"
            path: "pipelines/prompts/analysis/maintainability_focus.md"
          - type: "file"
            path: "src/main.py"
        output_to_file: "maintainability_analysis.json"

  - name: "synthesize_results"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/synthesis/comprehensive_synthesis.md"
      - type: "previous_response"
        step: "parallel_analysis"

Content Processing Features

Enhanced Extraction Options

prompt:
  - type: "previous_response"
    step: "code_analysis"
    extract_with: "content_extractor"  # Use ContentExtractor
    format: "structured"               # structured, summary, markdown
    post_processing:
      - "extract_code_blocks"
      - "extract_recommendations"
      - "extract_links"
    include_metadata: true
    max_length: 5000

Content Summarization

prompt:
  - type: "file"
    path: "large_specification.md"
    summary: true
    max_summary_length: 1000
  - type: "previous_response"
    step: "detailed_analysis"
    summary: true
    extract: "findings"

Variable Injection

prompt:
  - type: "file"
    path: "pipelines/prompts/analysis/project_analysis.md"
    variables:
      PROJECT_NAME: "MyApp"
      LANGUAGE: "Python"
      FRAMEWORK: "FastAPI"
  - type: "file"
    path: "${PROJECT_PATH}/src/main.py"
    inject_as: "SOURCE_CODE"

Best Practices

1. Prompt Organization

  • Single Responsibility: Each prompt file should focus on one specific task
  • Clear Naming: Use descriptive, action-oriented names
  • Version Control: Track prompt evolution with version comments
  • Documentation: Include purpose, context, and expected outputs

2. Template Design

  • Parameterization: Use variables for reusable templates
  • Flexibility: Design templates that work across different contexts
  • Clarity: Write clear, unambiguous instructions
  • Examples: Include examples of expected inputs and outputs

3. Component Architecture

  • Modularity: Design components that can be easily combined
  • Configuration: Support customization through variables
  • Reusability: Create components that work across workflows
  • Testing: Include test cases for component validation

4. Content Management

  • Caching: Leverage file caching for performance
  • Size Limits: Monitor and control prompt sizes
  • Processing: Use content extraction for large inputs
  • Validation: Verify prompt content before execution

5. Error Handling

  • Fallbacks: Provide fallback prompts for error conditions
  • Validation: Validate file paths and references
  • Recovery: Design recovery strategies for failed prompts
  • Monitoring: Track prompt performance and success rates

Complete Examples

Example 1: Full-Stack Application Analysis

workflow:
  name: "fullstack_app_analysis"
  
  steps:
    # Requirements gathering
    - name: "requirements_analysis"
      type: "gemini"
      prompt:
        - type: "file"
          path: "pipelines/prompts/analysis/requirements_analysis.md"
        - type: "file"
          path: "docs/requirements.md"
        - type: "file"
          path: "docs/user_stories.md"
      output_to_file: "requirements_analysis.json"

    # Architecture review
    - name: "architecture_review"
      type: "gemini"
      prompt:
        - type: "file"
          path: "pipelines/prompts/analysis/architecture_review.md"
        - type: "file"
          path: "docs/architecture.md"
        - type: "previous_response"
          step: "requirements_analysis"
          extract: "technical_requirements"
      output_to_file: "architecture_review.json"

    # Parallel code analysis
    - name: "code_analysis"
      type: "parallel_claude"
      parallel_tasks:
        - id: "frontend_analysis"
          claude_options:
            max_turns: 15
            allowed_tools: ["Read"]
          prompt:
            - type: "file"
              path: "pipelines/prompts/analysis/frontend_analysis.md"
            - type: "file"
              path: "frontend/src"
          output_to_file: "frontend_analysis.json"

        - id: "backend_analysis"
          claude_options:
            max_turns: 15
            allowed_tools: ["Read"]
          prompt:
            - type: "file"
              path: "pipelines/prompts/analysis/backend_analysis.md"
            - type: "file"
              path: "backend/src"
          output_to_file: "backend_analysis.json"

        - id: "database_analysis"
          claude_options:
            max_turns: 10
            allowed_tools: ["Read"]
          prompt:
            - type: "file"
              path: "pipelines/prompts/analysis/database_analysis.md"
            - type: "file"
              path: "database/schema.sql"
          output_to_file: "database_analysis.json"

    # Security assessment
    - name: "security_assessment"
      type: "claude_smart"
      preset: "analysis"
      prompt:
        - type: "file"
          path: "pipelines/prompts/validation/comprehensive_security_audit.md"
        - type: "previous_response"
          step: "code_analysis"
        - type: "previous_response"
          step: "architecture_review"
          extract: "security_considerations"
      output_to_file: "security_assessment.json"

    # Comprehensive report
    - name: "final_report"
      type: "gemini"
      token_budget:
        max_output_tokens: 8192
        temperature: 0.4
      prompt:
        - type: "file"
          path: "pipelines/prompts/generation/comprehensive_analysis_report.md"
        - type: "static"
          content: "## Requirements Analysis"
        - type: "previous_response"
          step: "requirements_analysis"
        - type: "static"
          content: "\n## Architecture Review"
        - type: "previous_response"
          step: "architecture_review"
        - type: "static"
          content: "\n## Code Analysis Results"
        - type: "previous_response"
          step: "code_analysis"
        - type: "static"
          content: "\n## Security Assessment"
        - type: "previous_response"
          step: "security_assessment"
      output_to_file: "comprehensive_analysis_report.md"

Example 2: Iterative Code Improvement

workflow:
  name: "iterative_code_improvement"
  
  steps:
    # Initial assessment
    - name: "initial_assessment"
      type: "gemini"
      prompt:
        - type: "file"
          path: "pipelines/prompts/analysis/code_quality_assessment.md"
        - type: "file"
          path: "src/legacy_code.py"
      functions:
        - "assess_code_quality"
      output_to_file: "initial_assessment.json"

    # First improvement iteration
    - name: "first_improvement"
      type: "claude_robust"
      retry_config:
        max_retries: 3
        backoff_strategy: "exponential"
      claude_options:
        max_turns: 20
        allowed_tools: ["Read", "Write", "Edit"]
      prompt:
        - type: "file"
          path: "pipelines/prompts/generation/code_improvement.md"
        - type: "previous_response"
          step: "initial_assessment"
          extract: "improvement_priorities"
        - type: "file"
          path: "src/legacy_code.py"
      output_to_file: "first_improvement.json"

    # Review first iteration
    - name: "review_first_iteration"
      type: "gemini"
      prompt:
        - type: "file"
          path: "pipelines/prompts/validation/improvement_review.md"
        - type: "previous_response"
          step: "first_improvement"
        - type: "previous_response"
          step: "initial_assessment"
          extract: "quality_targets"
      output_to_file: "first_review.json"

    # Second improvement iteration
    - name: "second_improvement"
      type: "claude_session"
      session_config:
        persist: true
        session_name: "code_improvement_session"
      prompt:
        - type: "claude_continue"
          new_prompt: |
            Based on the review feedback, please make the following additional improvements:
        - type: "previous_response"
          step: "review_first_iteration"
          extract: "additional_improvements"
      output_to_file: "second_improvement.json"

    # Final validation
    - name: "final_validation"
      type: "gemini"
      prompt:
        - type: "file"
          path: "pipelines/prompts/validation/final_quality_check.md"
        - type: "previous_response"
          step: "second_improvement"
        - type: "previous_response"
          step: "initial_assessment"
          extract: "quality_targets"
      functions:
        - "validate_improvements"
      output_to_file: "final_validation.json"

Migration Guide

Migrating from Inline Prompts

Before (Inline):

steps:
  - name: "analyze_code"
    type: "gemini"
    prompt:
      - type: "static"
        content: |
          Analyze this code for security issues:
          1. Check for SQL injection vulnerabilities
          2. Look for XSS vulnerabilities
          3. Review authentication mechanisms
          4. Check for insecure data handling
          
          Provide analysis in JSON format with severity levels.

After (File-based):

steps:
  - name: "analyze_code"
    type: "gemini"
    prompt:
      - type: "file"
        path: "pipelines/prompts/analysis/security_analysis.md"

Creating Prompt Libraries

  1. Extract Common Patterns: Identify frequently used prompt patterns
  2. Create Template Files: Move prompts to organized template files
  3. Add Variables: Parameterize templates for reusability
  4. Update Workflows: Convert workflows to use file references
  5. Test Migration: Verify equivalent functionality

Component Migration

  1. Identify Reusable Steps: Find step patterns used across workflows
  2. Create Component Definitions: Extract steps into component files
  3. Parameterize Components: Add variables for customization
  4. Update Workflows: Use component references instead of inline definitions
  5. Validate Components: Test components across different contexts

This guide provides a comprehensive framework for leveraging the Pipeline Prompt System's advanced capabilities. Use these patterns and standards to build maintainable, reusable, and powerful AI workflows.