Houtini-lm

# Houtini LM - Custom Prompting Guide *Master the art of flexible AI prompting with unlimited local processing* ## Overview The custom prompting function is your Swiss Army knife - a universal tool that handles any task when the specialized functions don't quite fit your needs. It's the bridge between Houtini's structured functions and your unique requirements. **Key Benefits**: - **Complete flexibility** - Any prompt, any task, any analysis - **Context-aware** - Automatically includes file contents when needed - **Token unlimited** - No API costs for experimentation - **Multi-file support** - Analyse entire projects or specific files ## The `custom_prompt` Function ### Basic Syntax ```bash local-llm:custom_prompt prompt="Your custom task description here" ``` ### Full Parameter Set | Parameter | Type | Required | Default | Description | |-----------|------|----------|---------|-------------| | `prompt` | string | Yes | - | Your custom instruction or question | | `filePath` | string | No | - | Single file to include as context | | `projectPath` | string | No | - | Project directory for multi-file context | | `files` | array | No | - | Specific files to include | | `code` | string | No | - | Code snippet to analyse | | `context` | object | No | {} | Additional structured context | | `analysisDepth` | string | No | "detailed" | Analysis depth: basic, detailed, comprehensive | | `maxDepth` | number | No | 3 | Directory traversal depth (1-5) | ## Prompt Engineering Best Practices ### 1. **Be Specific About Your Goal** ❌ Vague: "Look at this code" ✅ Specific: "Identify potential security vulnerabilities in this authentication system, focusing on input validation and session management" ### 2. **Specify Output Format** ❌ Generic: "Analyse this project" ✅ Structured: "Create a bulleted list of architectural improvements, ranked by impact and implementation difficulty" ### 3. **Include Context and Constraints** ❌ No context: "Optimize this function" ✅ With context: "Optimize this React component for a high-traffic e-commerce site, considering SEO and accessibility requirements" ### 4. **Request Confidence Scoring** Add phrases like: - "Rate your confidence in each recommendation from 1-10" - "Indicate which suggestions are high-priority vs nice-to-have" - "Mark any assumptions you're making about the codebase" ## Common Use Cases & Examples ### 1. **Custom Code Analysis** ```bash local-llm:custom_prompt prompt="Analyse this TypeScript codebase for potential performance bottlenecks, especially around database queries and API calls. Provide specific line numbers and optimization suggestions." projectPath="C:/my-api-project" ``` ### 2. **Architecture Assessment** ```bash local-llm:custom_prompt prompt="Review this microservices architecture and identify potential single points of failure. Consider scalability, fault tolerance, and monitoring gaps." projectPath="C:/microservices-app" analysisDepth="comprehensive" ``` ### 3. **Documentation Gap Analysis** ```bash local-llm:custom_prompt prompt="Identify which functions and classes lack proper documentation. Create a prioritised list based on code complexity and public API usage." filePath="C:/project/src/main-service.ts" ``` ### 4. **Framework Migration Planning** ```bash local-llm:custom_prompt prompt="Create a migration plan for converting this jQuery application to React. Identify reusable components, potential challenges, and step-by-step conversion approach." projectPath="C:/legacy-webapp" ``` ### 5. **Custom Security Audit** ```bash local-llm:custom_prompt prompt="Perform a WordPress-specific security audit focusing on custom post types, meta queries, and user capability checks. Flag any potential privilege escalation vectors." projectPath="C:/wp-plugin" ``` ### 6. **Performance Analysis** ```bash local-llm:custom_prompt prompt="Analyse this React application for unnecessary re-renders and prop drilling. Suggest state management improvements and component optimization strategies." projectPath="C:/react-dashboard" ``` ### 7. **Code Style Consistency** ```bash local-llm:custom_prompt prompt="Review coding standards consistency across this team project. Identify style variations, naming convention deviations, and suggest linting rules." projectPath="C:/team-project" maxDepth=4 ``` ### 8. **API Design Review** ```bash local-llm:custom_prompt prompt="Evaluate this REST API design for RESTful principles, error handling consistency, and developer experience. Suggest improvements for API versioning and documentation." filePath="C:/api/routes/users.js" ``` ## Advanced Prompting Techniques ### 1. **Multi-Step Analysis** Break complex tasks into steps within your prompt: ```bash local-llm:custom_prompt prompt=" Step 1: Identify all database queries in this codebase Step 2: Categorise them by complexity and frequency of execution Step 3: Suggest specific optimisations for the top 5 slowest queries Step 4: Estimate performance impact of each optimisation " projectPath="C:/database-heavy-app" ``` ### 2. **Role-Based Prompting** Frame your request from a specific perspective: ```bash local-llm:custom_prompt prompt="As a senior DevOps engineer, review this application's deployment configuration. Focus on container security, resource limits, and CI/CD pipeline improvements. What would you change before deploying to production?" projectPath="C:/app-deployment" ``` ### 3. **Comparative Analysis** Ask for comparisons and trade-offs: ```bash local-llm:custom_prompt prompt="Compare the current authentication implementation with OAuth 2.0 and JWT approaches. Create a pros/cons table focusing on security, scalability, and development complexity." filePath="C:/auth/current-auth.js" ``` ### 4. **Scenario-Based Analysis** Present specific scenarios: ```bash local-llm:custom_prompt prompt="If this application needs to handle 10x current traffic within 6 months, what are the top 5 architectural changes needed? Consider database scaling, caching strategies, and infrastructure requirements." projectPath="C:/startup-app" ``` ## Working with Different File Types ### **JavaScript/TypeScript Projects** ```bash local-llm:custom_prompt prompt="Analyse this Node.js API for async/await best practices, error handling patterns, and potential memory leaks" projectPath="C:/node-api" ``` ### **WordPress/PHP Projects** ```bash local-llm:custom_prompt prompt="Review this WordPress plugin for compliance with coding standards, security best practices, and performance optimization opportunities" projectPath="C:/wp-plugin" ``` ### **Configuration Files** ```bash local-llm:custom_prompt prompt="Audit this Docker Compose configuration for production readiness, security hardening, and performance optimization" filePath="C:/project/docker-compose.yml" ``` ### **Documentation Files** ```bash local-llm:custom_prompt prompt="Review this API documentation for completeness, clarity, and developer experience. Identify missing examples, unclear explanations, and suggest improvements" filePath="C:/docs/api-guide.md" ``` ## Context Enhancement Strategies ### 1. **Structured Context Objects** Provide additional context for better analysis: ```bash local-llm:custom_prompt prompt="Optimise this component for mobile performance" filePath="C:/components/DataTable.tsx" context='{ "target_devices": "iOS/Android", "performance_budget": "< 3s load time", "constraints": ["limited bandwidth", "touch interface"], "framework": "React Native" }' ``` ### 2. **Project-Specific Requirements** Include business context: ```bash local-llm:custom_prompt prompt="Review this e-commerce checkout flow for conversion optimization" projectPath="C:/checkout-app" context='{ "business_goals": ["reduce cart abandonment", "increase average order"], "target_audience": "mobile-first shoppers", "compliance": ["PCI DSS", "GDPR"] }' ``` ## Output Formatting Techniques ### 1. **Request Specific Formats** ```bash local-llm:custom_prompt prompt="Create a markdown checklist of security improvements for this API, grouped by priority level (Critical, High, Medium, Low)" filePath="C:/api/auth.js" ``` ### 2. **Ask for Templates** ```bash local-llm:custom_prompt prompt="Generate a pull request template for this project based on the codebase structure and apparent coding standards" projectPath="C:/team-project" ``` ### 3. **Request Actionable Outputs** ```bash local-llm:custom_prompt prompt="Create a step-by-step refactoring plan with estimated time requirements and risk levels for modernising this legacy component" filePath="C:/legacy/OldWidget.js" ``` ## Troubleshooting Common Issues ### **Vague or Unhelpful Responses** - **Make prompts more specific** - Add context about your goals - **Include examples** of what you're looking for - **Specify the output format** you need ### **Analysis Too Surface-Level** - **Increase `analysisDepth`** to "comprehensive" - **Add specific technical constraints** to your prompt - **Request line-by-line analysis** for critical files ### **Missing Important Files** - **Use `files` parameter** to specify exactly which files to analyse - **Increase `maxDepth`** for complex project structures - **Combine multiple calls** for very large projects ### **Context Not Relevant** - **Be specific about your role** and requirements - **Include business constraints** and technical requirements - **Mention the target audience** for your analysis ## Integration with Other Functions ### **Use Custom Prompt to Plan** 1. `custom_prompt` - "What security issues should I look for in this API?" 2. `security_audit` - Comprehensive automated audit 3. `custom_prompt` - "Create implementation plan for fixing the identified issues" ### **Use Custom Prompt to Refine** 1. `analyze_single_file` - Get structured analysis 2. `custom_prompt` - "Focus specifically on the performance concerns from the previous analysis" 3. `suggest_refactoring` - Get specific improvement suggestions ### **Use Custom Prompt for Follow-up** 1. `generate_unit_tests` - Create test suite 2. `custom_prompt` - "Review these generated tests for edge cases I might have missed" ## Performance Optimization ### **For Large Projects** - Start with `maxDepth=2` to get overview, then drill down - Use specific file paths rather than entire projects when possible - Break complex analysis into multiple focused prompts ### **For Better Response Quality** - Use `analysisDepth="comprehensive"` for critical analysis - Include relevant technical constraints in your context - Ask for confidence levels and assumptions in responses ### **For Faster Iteration** - Save effective prompts as templates - Build up context gradually rather than trying to include everything at once - Use the structured `context` parameter for consistent formatting ## Best Practices Summary ### **Do** ✅ Be specific about your goals and constraints ✅ Include relevant business or technical context ✅ Request structured outputs with confidence levels ✅ Break complex analysis into focused questions ✅ Specify output formats (markdown, lists, tables) ✅ Include file paths when relevant to the analysis ### **Don't** ❌ Use vague prompts like "analyse this code" ❌ Try to analyse massive projects in single calls ❌ Forget to specify the technical stack or framework ❌ Assume the AI knows your business requirements ❌ Skip context about performance requirements or constraints --- ## Examples Repository For more examples and templates, check out our [examples repository](https://github.com/houtini-ai/examples) with real-world custom prompting scenarios. *The custom_prompt function is your gateway to unlimited analysis possibilities. Master it, and you'll be able to tackle any development challenge with AI assistance.*