Skip to content

TODO.md

Latest commit

 

History

History
143 lines (105 loc) · 7.48 KB

TODO.md

File metadata and controls

143 lines (105 loc) · 7.48 KB
Raw

Octave — Development Todo List

Project Foundation

  • 1. Initialize Python backend project structure (virtual env, dependencies, FastAPI/Flask server)
  • 2. Initialize React frontend project structure (Vite/Create React App, routing setup)
  • 3. Establish backend-frontend communication layer (REST API + WebSocket for real-time chat)
  • 4. Set up vector-capable database schema and ORM (SQLite with vec0 extension; pgvector deferred to a pluggable adapter; Alembic for migrations) — PR #84

UI — Core Layout

  • 1. Build main application shell (navigation, sidebar, content area)
  • 2. Create responsive layout components and shared UI component library

UI — Chat Interface

  • 1. Build chat message display component (user/agent messages, timestamps)
  • 2. Implement chat input component (text area, send button, attachment support)
  • 3. Create real-time message streaming (SSE/WebSocket integration for agent responses)
  • 4. Add conversation history management (list, switch, delete conversations)

UI — MCP Connector View

  • 1. Build MCP server list view (connected servers, status indicators)
  • 2. Create MCP server add/edit/remove UI (configuration form, connection testing)
  • 3. Build MCP tool explorer (browse available tools per server, view tool schemas)
  • 4. Add MCP tool invocation interface (manual tool testing/debugging UI)

UI — Context Manager View

  • 1. Build context vault browser (skills, prompts, preferences categories)
  • 2. Create context item editor (CRUD for skills, prompts, preferences)
  • 3. Build agent context viewer (current running context, injected items)

UI — Agent Manager View

  • 1. Build agent status dashboard (list running agents, states, progress indicators)
  • 2. Add agent panel to chat interface (show active agent context alongside conversation)
  • 3. Create agent result viewer (browse outputs from completed agent runs)

UI — Settings

  • 1. Create inference engine settings panel (endpoint URL, API key, model selection)
  • 2. Build MCP global settings (default transport, timeout, retry policy)
  • 3. Create user preferences panel (theme, language, default models)

Inference Engine Connector

  • 1. Design pluggable inference engine interface (abstract base class for engine adapters)
  • 2. Implement generic REST-based inference adapter (OpenAI-compatible API format)
  • 3. Add streaming response support for real-time token output
  • 4. Build prompt assembly pipeline (receives assembled context from Context Manager, constructs final prompt for inference)
  • 5. Add engine health check and fallback mechanism
  • 6. Add embedding model support (generate vector embeddings for context indexing and semantic search)
  • 7. Implement model tagging system (label models with capability tags, e.g., "thinking", "coding", "quick")

MCP Connector

  • 1. Implement MCP client core (JSON-RPC 2.0 transport layer)
  • 2. Add MCP stdio transport support (spawn subprocess servers) — PR #81 (transport behind open_transport) + PR #91 (subprocess exit capture, manual restart); config persistence + Settings UI wiring deferred to #7/UI
  • 3. Add MCP HTTP/SSE transport support (remote server connections) — reduced by PR #81 to config persistence + Settings UI wiring; Streamable HTTP exists behind open_transport (legacy SSE deliberately not wrapped)
  • 4. Build server lifecycle manager (start, stop, restart, health monitoring) — PR #93 (per-server supervisors, backoff + crash-loop, probe-on-timeout; REST surface deferred to MCP Connector UI)
  • 5. Implement tool discovery and caching (fetch tools, schemas, descriptions) — PR #98 (ToolRegistry inventory; event-driven invalidation; cursor pagination deferred to follow-up issue)
  • 6. Create tool execution engine (invoke tools, handle responses/errors, timeouts) — PR #98 ((server_id, tool_name) surface over McpClient.call_tool)
  • 7. Add MCP server configuration persistence (store server configs in the unified database)
  • 8. Implement MCP tool tagging system (label tools for internal Octave use, e.g., context retrieval, file operations)
  • 9. Add tool re-naming support (map custom names to underlying MCP tool names for agent-facing clarity)
  • 10. Add tool re-describing support (override tool descriptions for agents while preserving original MCP mapping)

Context Manager

  • 1. Design context vault data model (skills, prompts, preferences, agent state schemas)
  • 2. Implement vector-capable storage layer for context vault (CRUD operations, queries, vector indexing) — PR #97
  • 3. Build context vault using tagged MCP tools (discover and invoke tagged tools to populate skills, prompts, and preferences into the vault) [Depends on: MCP Connector #8–10]
  • 4. Create context injection engine (select relevant context items based on rules/triggers)
  • 5. Implement agent context lifecycle (receive full context from completed agent runs via Agent Manager, embed into vector DB for future linked agent runs to query) [Depends on: Agent Manager #1–4]
  • 6. Add context relevance scoring or filtering (token budget management, priority ranking)
  • 7. Build external database adapter interface (optional MCP-provided database override)
  • 8. Implement skill-to-tool/prompt linking (associate skills with specific tools or prompts)
  • 9. Add skill parameter templating (auto-replace placeholder parameters in skills with linked tool/prompt names)
  • 10. Add conversation-to-vector indexing (convert conversation messages into embeddings stored in the vector DB for semantic search)
  • 11. Build vector search query interface (allow agents and context manager to perform semantic similarity search across conversations and vault items)
  • 12. Add skill-to-model linking via tags (attach required model tags to skills; resolve to matching models at runtime)

Agent Manager

  • 1. Design agent lifecycle model (definition active/paused gate; instance spawn → idle ⇄ active → destroy) — PR #105 (lifecycle vocabulary, agent_instances table, AgentInstanceManager; router/registry/scheduling remain #2–6)
  • 2. Build agent registry (track running agents, their IDs, status, and assigned context)
  • 3. Implement agent message routing (deliver messages to correct agent, broadcast when needed)
  • 4. Create agent result collection (capture agent outputs and make them queryable)
  • 5. Build inter-agent result sharing (allow agents to request and receive results from other agents)
  • 6. Add agent priority and scheduling (queue management, resource constraints)

Integration & Testing

  • 1. Wire end-to-end flow: User message → Context assembly → Inference → MCP tool calls → Response
  • 2. Create integration tests for inference connector
  • 3. Create integration tests for MCP connector (mock servers)
  • 4. Create integration tests for context injection pipeline
  • 5. Build basic smoke tests for UI components
  • 6. Create integration tests for agent manager (multi-agent workflows)

Summary

Area Items
Project Foundation 4
UI — Core Layout 2
UI — Chat Interface 4
UI — MCP Connector View 4
UI — Context Manager View 3
UI — Settings 3
UI — Agent Manager View 3
Inference Engine Connector 7
MCP Connector 10
Context Manager 12
Agent Manager 6
Integration & Testing 6
Total 64