SDOF Knowledge Base

+++ id = "ROO-CMD-RULE-SDOF-RAG-INTEGRATION-V1" title = "Roo Commander: Workspace-Wide Rule for SDOF & RAG Integration" status = "active" scope = "Workspace-wide rule providing reusable instructions for dynamic SDOF context retrieval, prompt caching, and token management for all modes" target_audience = ["all"] tags = ["sdof", "rag", "context-retrieval", "prompt-caching", "token-management", "workspace-wide", "integration"] related_docs = [ ".roo/rules/rules.md", ".ruru/docs/guides/06_Advanced_Usage_Customization/02_Custom_Instructions_Rules.md", ".ruru/docs/architecture/sdof_architecture_design_v1.md", ".ruru/docs/api/sdof_knowledge_system_v1.md" ] +++ # Workspace-Wide Rule: SDOF & RAG Integration ## Objective This workspace-wide rule defines reusable, standardized instructions for all modes to dynamically retrieve relevant context from the SDOF Knowledge Base using MCP tools, implement prompt caching strategies based on `cache_hint` metadata, and monitor token usage with context pruning to optimize LLM API requests. The goal is to provide a consistent, maintainable integration of advanced AI capabilities and SDOF knowledge retrieval across all modes in the workspace. ## Instructions for Modes ### 1. Dynamic SDOF Context Retrieval - Analyze the user query or task to identify key entities, concepts, and keywords. - Use the prioritized retrieval strategy as defined in the project rules ([`.roo/rules/rules.md`](.roo/rules/rules.md#dynamic_knowledge_retrieval)) to select appropriate SDOF MCP tools: - Targeted knowledge search using `store_sdof_plan` with query parameters. - Specific knowledge retrieval by planType, phase, or tags. - Contextual expansion via related knowledge links. - Fallback to broad context retrieval for general project knowledge. - Synthesize and filter retrieved knowledge to produce concise, relevant summaries. - Handle errors gracefully, logging and falling back as needed. ### 2. Prompt Caching Strategy - Identify cacheable content from retrieved SDOF knowledge based on: - Presence of `cache_hint: true` in `metadata` of stored plans. - Size and stability heuristics (e.g., minimum token threshold). - Prioritized plan types (`project_context`, `system_pattern`, `implementation_knowledge`, `decision_record`). - Structure prompts for caching according to the LLM provider: - For implicit caching (e.g., Gemini, OpenAI), place stable SDOF knowledge at the absolute beginning of the prompt. - For explicit caching (e.g., Anthropic), insert `cache_control` breakpoints after stable SDOF content. - Notify users when prompt caching is applied: `[INFO: Structuring prompt for caching]`. ### 3. Token Usage Monitoring and Context Pruning - Estimate token usage of the constructed prompt before sending to the LLM. - If token usage exceeds limits, prune context by: - Removing less relevant or lower priority SDOF knowledge items. - Summarizing or compressing SDOF knowledge where possible. - Ensure the final prompt respects token limits while maintaining relevance. ### 4. Integration and Usage - Modes should load this rule automatically as it is workspace-wide. - Modes can invoke the instructions defined here to fetch SDOF knowledge and build optimized prompts. - This rule complements mode-specific system prompts and rules, providing advanced context management capabilities. ### 5. SDOF Knowledge Base Operations #### Storage Operations - **Trigger**: When new insights, decisions, implementations, or learnings emerge - **Tool**: `store_sdof_plan` with structured metadata - **Format**: Markdown content with comprehensive metadata including: - `planTitle`: Descriptive title for the knowledge entry - `planType`: Category (e.g., "decision_record", "implementation_note", "system_pattern") - `tags`: Relevant keywords for retrieval - `phase`: SDOF phase if applicable (1-5) - `cache_hint`: Boolean indicating high-value caching content #### Retrieval Operations - **Trigger**: When needing project context, previous decisions, or implementation details - **Tool**: `store_sdof_plan` with query parameters (when available) or direct retrieval - **Strategy**: - Query by specific tags for targeted knowledge - Filter by planType for category-specific knowledge - Retrieve by phase for workflow-specific context - Use semantic search capabilities when available #### Error Handling - Gracefully handle SDOF Knowledge Base service unavailability - Provide fallback behavior when specific knowledge isn't found - Log retrieval attempts and failures for debugging - Continue operation in limited mode when SDOF is unavailable ## References - Project rules on dynamic knowledge retrieval and RAG: [`.roo/rules/rules.md`](.roo/rules/rules.md#dynamic_knowledge_retrieval) - Prompt caching strategies: [`.roo/rules/rules.md`](.roo/rules/rules.md#prompt_caching_strategies) - Advanced usage guide: [`.ruru/docs/guides/06_Advanced_Usage_Customization/02_Custom_Instructions_Rules.md`](.ruru/docs/guides/06_Advanced_Usage_Customization/02_Custom_Instructions_Rules.md) - SDOF Architecture design: [`.ruru/docs/architecture/sdof_architecture_design_v1.md`](.ruru/docs/architecture/sdof_architecture_design_v1.md) - SDOF Knowledge system: [`.ruru/docs/api/sdof_knowledge_system_v1.md`](.ruru/docs/api/sdof_knowledge_system_v1.md) ## Change Log - v1 (2025-06-01): Initial version migrated from ConPort to SDOF, defining workspace-wide SDOF and RAG integration instructions.