Working on LocalPilot

Commands

• Type-check + lint + bundle: npm run compile
• Offline logic tests (mock Ollama/OpenAI servers, agent loop): npm run smoke
• Launch the extension host: VS Code Run Extension (F5)

Architecture map

Area	Path	Notes
Model backends	src/providers/	ollama.ts (native API), openaiCompat.ts (LM Studio, llama.cpp, vLLM…), registry.ts resolves provider/model ids
Agent loop	src/agent/agentLoop.ts	model ⇄ tools loop, parallel tool execution, text-protocol fallback
Subagents & skills	src/agent/subagents.ts, src/agent/skills.ts	run_subagent, run_parallel_subagents, load_skill
Context window	src/context/contextManager.ts	prune → summarise → truncate
Tools	src/tools/*.ts	each exports Tool[]; register in src/extension.ts
UI	src/ui/, media/chat.{js,css}	webview controller, chat participant, LM provider, inline completions
MCP	src/mcp/mcpManager.ts	stdio JSON-RPC client

Adding a tool

1. Create src/tools/<name>.ts exporting a Tool with a JSON-schema definition.
2. Mark parallelSafe: true if read-only; set approvalKind for anything mutating.
3. Add it to the registerAll call in src/extension.ts.
4. Add a toolSummary case in media/chat.js for a nice label.

Adding a provider type

Implement LLMProvider from src/types.ts, then branch on cfg.type in createProvider() in src/providers/registry.ts and extend the enum in package.json → localpilot.providers.