# {{Agent Name}} LLM Agent Specification [[LLM: Dynamic Specification Generation 1. **Context Analysis**: Understand the specific project context, requirements, and constraints 2. **Research Current Standards**: Research latest best practices for LLM agent specifications 3. **Adaptive Template**: Customize this template based on analyzed needs and current industry standards 4. **Technology-Aware**: Consider current LLM capabilities and limitations when setting requirements 5. **Requirements Elicitation**: Use systematic elicitation to uncover both explicit and implicit needs 6. **Future-Proofing**: Design specifications that can evolve with advancing LLM capabilities **Research Before Starting**: - What are current best practices for LLM agent specifications in this domain? - What recent developments in LLM should influence the design requirements? - What are the latest safety and compliance considerations? - How do successful similar agents approach the specified use case? Output file location: `docs/llm-agents/{{agent-name}}-spec.md`]] ## Agent Overview [[LLM: Provide a comprehensive overview of the LLM agent, including its purpose, target users, and value proposition. Be specific about the problems it solves.]] - **Agent Name**: {{agent-name}} - **Agent Type**: {{Conversational/Task-oriented/Autonomous/Multi-modal}} - **Primary Purpose**: {{main-purpose}} - **Target Users**: {{user-segments}} - **Key Capabilities**: {{list 3-5 core capabilities}} ## Functional Requirements [[LLM: Detail what the agent must do. Be specific about inputs, outputs, and behaviors.]] ### Core Capabilities <> #### {{Capability Name}} - **Description**: {{detailed-description}} - **Input Types**: {{expected-inputs}} - **Output Format**: {{output-structure}} - **Success Criteria**: {{measurable-criteria}} - **Error Handling**: {{error-scenarios}} <> ### Conversation Flow [[LLM: For conversational agents, define the interaction patterns and dialogue management.]] - **Greeting/Initialization**: {{initial-interaction}} - **Context Management**: {{how-context-is-maintained}} - **Turn-Taking**: {{conversation-rules}} - **Closure/Handoff**: {{end-of-interaction}} ## Non-Functional Requirements ### Performance Requirements [[LLM: Define specific, measurable performance targets.]] - **Response Time**: {{latency-requirements}} - **Throughput**: {{requests-per-second}} - **Availability**: {{uptime-percentage}} - **Scalability**: {{scaling-requirements}} - **Resource Limits**: {{cpu/memory/cost-constraints}} ### Safety and Alignment [[LLM: Detail safety measures and alignment strategies.]] - **Content Filtering**: {{harmful-content-prevention}} - **Bias Mitigation**: {{fairness-measures}} - **Privacy Protection**: {{data-handling-policies}} - **Security Measures**: {{injection-prevention}} - **Alignment Constraints**: {{behavioral-boundaries}} ## Technical Architecture ### Model Selection [[LLM: Research and document model choices with current analysis and rationale.]] **Research Process**: - Research current available models and their capabilities for this use case - Analyze performance benchmarks relevant to the agent's requirements - Evaluate cost implications across different model options - Consider integration complexity and API stability **Model Strategy**: - **Primary Model**: {{model-name-version}} - Selected based on {{research-based-rationale}} - **Fallback Models**: {{backup-options}} - Chosen for {{fallback-reasoning}} - **Selection Criteria**: {{research-informed-criteria}} - **Fine-tuning Requirements**: {{customization-needs-analysis}} - **Cost Considerations**: {{budget-constraints-and-optimization}} ### Integration Points [[LLM: Define how the agent connects with other systems.]] - **APIs/Services**: {{external-dependencies}} - **Data Sources**: {{knowledge-bases}} - **Authentication**: {{auth-methods}} - **Rate Limiting**: {{throttling-policies}} - **Error Recovery**: {{retry-strategies}} ## Prompt Engineering Strategy ### System Prompts [[LLM: Define the core system prompts that shape agent behavior.]] ```text {{system-prompt-template}} ``` ### Prompt Patterns - **Task Decomposition**: {{how-complex-tasks-are-handled}} - **Context Injection**: {{dynamic-context-management}} - **Output Formatting**: {{structured-response-patterns}} - **Error Handling**: {{graceful-failure-prompts}} ### Testing Strategy [[LLM: Research current testing methodologies and design appropriate testing strategy.]] **Research Areas**: - Current best practices for LLM agent testing and evaluation - Available testing frameworks and their suitability for this use case - Industry-standard evaluation metrics for similar agents - Latest developments in prompt testing and optimization **Testing Approach**: - **Test Coverage**: {{research-informed-test-scenarios}} - **Evaluation Metrics**: {{current-best-practice-measures}} - **Testing Framework**: {{selected-framework-with-rationale}} - **Optimization Strategy**: {{evidence-based-optimization-approach}} ## Monitoring and Observability ### Key Metrics [[LLM: Define what will be measured and monitored.]] - **Quality Metrics**: {{accuracy/relevance/completeness}} - **Performance Metrics**: {{latency/throughput/errors}} - **Business Metrics**: {{user-satisfaction/engagement}} - **Cost Metrics**: {{per-request/daily-budget}} ### Alerting Thresholds <> - **{{Metric Name}}**: - Warning: {{warning-threshold}} - Critical: {{critical-threshold}} - Action: {{remediation-steps}} <> ## Development Phases ### Phase 1: MVP - **Timeline**: {{duration}} - **Features**: {{core-features-only}} - **Success Criteria**: {{mvp-goals}} ### Phase 2: Enhancement - **Timeline**: {{duration}} - **Features**: {{additional-capabilities}} - **Success Criteria**: {{enhancement-goals}} ### Phase 3: Scale - **Timeline**: {{duration}} - **Features**: {{scale-features}} - **Success Criteria**: {{scale-goals}} ## Risk Assessment [[LLM: Identify and plan for potential risks.]] <> ### {{Risk Name}} - **Probability**: {{High/Medium/Low}} - **Impact**: {{High/Medium/Low}} - **Mitigation**: {{prevention-strategy}} - **Contingency**: {{if-it-happens}} <> ## Testing and Validation ### Test Scenarios [[LLM: Define comprehensive test cases covering normal and edge cases.]] 1. **Functional Tests**: {{core-functionality-tests}} 2. **Edge Cases**: {{boundary-conditions}} 3. **Adversarial Tests**: {{security-tests}} 4. **Performance Tests**: {{load-testing}} 5. **Integration Tests**: {{system-integration}} ### Acceptance Criteria - [ ] All functional requirements met - [ ] Performance targets achieved - [ ] Safety measures validated - [ ] Integration tests passing - [ ] User acceptance confirmed ## Documentation Requirements - **API Documentation**: {{api-spec-location}} - **User Guide**: {{end-user-docs}} - **Operations Manual**: {{ops-runbook}} - **Integration Guide**: {{developer-docs}} ## Compliance and Legal - **Data Protection**: {{gdpr/ccpa-compliance}} - **Industry Standards**: {{relevant-standards}} - **Audit Requirements**: {{logging-needs}} - **Terms of Service**: {{usage-policies}} ## Success Metrics [[LLM: Define how success will be measured post-deployment.]] - **Adoption Metrics**: {{user-growth-targets}} - **Quality Metrics**: {{satisfaction-scores}} - **Business Metrics**: {{roi-targets}} - **Technical Metrics**: {{reliability-goals}} ## Next Steps 1. Review and approve specification 2. Create detailed technical design 3. Set up development environment 4. Implement MVP features 5. Configure testing framework 6. Plan deployment strategy