memory-bank-vscode — Copilot Instructions

Core Memory Bank (Semantic RAG) | Tool | When to Use | |------|-------------| | memorybank_route_task | FIRST - Before ANY code changes | | memorybank_search | SECOND - Before implementation | | memorybank_index_code | AFTER any modification | | memorybank_read_file | When need full file context | | memorybank_write_file | Write with auto-reindex | Multi-Project | Tool | Description | |------|-------------| | memorybank_manage_agents | Coordination, locking & task management | | memorybank_discover_projects | Find other projects | | memorybank_delegate_task | Handoff work to other projects | Agent Board Actions (memorybank_manage_agents) | Action | Description | |--------|-------------| | register | Register agent at session start | | get_board | View agents, tasks, locks | | claim_task | Claim a pending task | | complete_task | Mark task as completed | | claim_resource | Lock a file/resource | | release_resource | Unlock a file/resource | | update_status | Update agent status | Project Knowledge Layer | Tool | Description | |------|-------------| | memorybank_generate_project_docs | Generate AI docs (replaces templates) | | memorybank_get_project_docs | Read project documentation | Context Management | Tool | Description | |------|-------------| | memorybank_initialize | Create basic templates (no AI, instant) | | memorybank_update_context | Update session context | | memorybank_record_decision | Record decisions | | memorybank_track_progress | Track progress | | memorybank_manage_agents | Coordination & locking | MCP Resources | Resource URI | Content | |--------------|---------| | memory://memory_bank_vscode_extension/active | Session context | | memory://memory_bank_vscode_extension/progress | Progress | | memory://memory_bank_vscode_extension/decisions | Decisions | | memory://memory_bank_vscode_extension/context | Project context | ---

• Project ID: memory_bank_vscode_extension • Mode: Auto-Index (continuous RAG synchronization) ---

This project uses Memory Bank MCP as a RAG system (Retrieval-Augmented Generation). It keeps your knowledge accurate and prevents hallucinations.

Rule 0: COORDINATE WITH OTHER AGENTS BEFORE starting any task, you MUST check the Agent Board. This prevents multiple agents from modifying the same files simultaneously or duplicating work. • Check Board: Use memorybank_manage_agents with action: "get_board" to see active agents/locks. • Register: Identity yourself (e.g., role-ide-model). Call action: "register" with your agentId. The system will assign a Session ID for tracking context automatically. • Claim Task: action: "claim_resource" for the file/feature you are working on. • Work: Perform your task (Route -> Search -> Implement -> Index). • Release: action: "release_resource" when done. Rule 0.5: ROUTE BEFORE IMPLEMENTING (MANDATORY) BEFORE writing ANY code, you MUST call memorybank_route_task to determine what belongs to this project vs what should be delegated. `json // memorybank_route_task - MANDATORY before ANY code changes { "projectId": "memory_bank_vscode_extension", "taskDescription": "[describe what you're about to implement]" } ` The orchestrator will analyze the task against ALL projects' responsibilities and return: • myResponsibilities: What YOU should implement in this project • delegations: What should be delegated to other projects • suggestedImports: What to import from other projects • architectureNotes: Design guidance Why this matters: • Prevents creating code that belongs to other projects (e.g., DTOs in API when lib-dtos exists) • Ensures proper separation of concerns across the ecosystem • Automatically delegates work to the right project NO EXCEPTIONS. Even for "simple" tasks - the orchestrator knows the ecosystem better than you. Rule 1: ALWAYS SEARCH BEFORE IMPLEMENTING NEVER write code without first consulting the Memory Bank. `json // memorybank_search - MANDATORY before ANY implementation { "projectId": "memory_bank_vscode_extension", "query": "how does [feature/component] work" } ` When to search: • Before implementing anything → Search for similar patterns • Before modifying code → Search for usages and dependencies • Before answering questions → Search for accurate info • Before suggesting architecture → Search for existing patterns Rule 2: ALWAYS REINDEX AFTER MODIFYING IMMEDIATELY after modifying ANY file, reindex it. `json // memorybank_index_code - MANDATORY after ANY file change { "projectId": "memory_bank_vscode_extension", "path": "path/to/modified/file.ts" } ` No exceptions. Keeps the RAG updated and accurate. Rule 3: RESPECT PROJECT BOUNDARIES You own memory_bank_vscode_extension. Do NOT modify other projects. • Discover: memorybank_discover_projects to find other agents. • Delegate: memorybank_delegate_task to hand off work. Rule 4: DOCUMENT EVERYTHING CONTINUOUSLY After EVERY significant action, update the Memory Bank: • Track progress after completing ANY task: `json { "projectId": "memory_bank_vscode_extension", "progress": { "completed": ["Task done"], "inProgress": ["Next task"] } } ` • Record decisions when making architectural/technical choices: `json { "projectId": "memory_bank_vscode_extension", "decision": { "title": "...", "description": "...", "rationale": "..." } } ` • Update context to leave notes for next session: `json { "projectId": "memory_bank_vscode_extension", "recentChanges": ["..."], "nextSteps": ["..."] } ` The goal: Next session (or another agent) can pick up exactly where you left off. ---