v0.6.0

Breaking

• src/chats/ split into src/chat/ and src/service/; src/tools/base.ts split into tool.ts, parameter.ts, result.ts, package.ts, hook.ts — update deep imports
• executeTool() return type changed from {result, status}[] to ToolResult[] (adds tool and optional error fields)

Features

• reasoningEffort config (None–XHigh), env LLM_CHAT_OPENAI_REASONING_EFFORT
• toolChoice config (None/Auto/Required), env LLM_CHAT_OPENAI_TOOL_CHOICE
• verbosity config (Low/Medium/High), env LLM_CHAT_OPENAI_VERBOSITY
• ChatRole.Developer; useDeveloperRole flag maps system messages to role: developer
• systemPrompt config string bypasses the prompt tree
• trimMessages option strips whitespace from assistant/reasoning content
• ToolParameterProperty sub-schemas + ObjectPropertyBuilder; recursive toJSON() for properties, required, and custom array items
• beforeSendLoop/afterSendLoop/beforeSend/afterSend service hooks via service.hook(), reentrant-safe
• ToolSuite.get(name); executeTool(name, args, silent); ChatService.injectToolCall()
• reasoning_details/reasoning/reasoning_content extraction from stream deltas, priority cascade
• Service, message, and tool hooks await async callbacks

Internal

• Stream types extracted to src/service/stream-types.ts
• _drainQueue() extracted — 4 inline drain blocks in send() consolidated
• _pendingReentrant cleared each iteration, fixing infinite loop
• ChatMessageOrigin.Hook guard prevents infinite JSON hook→queue→hook loops
• Added p-map and openai dependencies
• Documentation updated

Full Changelog: v0.5.0...v0.6.0