Kokorone/apps/api/src/llm/dialogue.service.ts

import { Inject, Injectable } from "@nestjs/common";
import { PrismaService } from "../prisma/prisma.service.js";
import { WorkingMemoryService } from "../memory/working-memory.service.js";
import { EmotionService } from "../emotion/emotion.service.js";
import { RelationshipService } from "../relationship/relationship.service.js";
import { LLM_PROVIDER, type LLMProvider } from "./llm-provider.js";
import { ContextAssemblerService } from "./context-assembler.service.js";
import { OutputFilterService } from "./output-filter.service.js";
import { FastChannelDetector, DEFAULT_FAST_REPLIES } from "./fast-channel.service.js";
import { extractActions } from "./action-markup.js";
import { getForbiddenWords, applySpeechQuirks } from "../personality/language-style.js";
import type { GenerationContext } from "./types.js";

export interface HandleMessageOptions {
  seed?: number;
  now?: Date;
}

export interface HandleMessageResult {
  text: string;
  actions: string[];
  isFastChannel: boolean;
  context: GenerationContext;
}

function fastChannelSituation(reason: string): string {
  return `fast-channel:${reason}`;
}

// F 群組整合入口：串起雙速通道判斷、上下文組裝、Provider 生成、輸出過濾與工作記憶寫入。
// H-1 的 POST /chat/:characterId 會以此為基礎擴充（見 todo.md 實作記錄）。
@Injectable()
export class DialogueService {
  constructor(
    @Inject(LLM_PROVIDER) private readonly llmProvider: LLMProvider,
    private readonly fastChannel: FastChannelDetector,
    private readonly contextAssembler: ContextAssemblerService,
    private readonly outputFilter: OutputFilterService,
    private readonly workingMemory: WorkingMemoryService,
    private readonly emotion: EmotionService,
    private readonly relationship: RelationshipService,
    private readonly prisma: PrismaService,
  ) {}

  async handleMessage(
    characterId: string,
    userId: string,
    sessionId: string,
    userInput: string,
    options: HandleMessageOptions = {},
  ): Promise<HandleMessageResult> {
    const now = options.now ?? new Date();
    this.workingMemory.append(sessionId, { role: "user", content: userInput, timestamp: now });

    // H-1：每輪對話先標記情緒、累積日常互動的親密度，讓後續組裝的上下文反映本輪的變化。
    await this.emotion.processInput(characterId, userInput, { now });
    await this.relationship.recordInteraction(characterId, userId, now);

    const { isFastChannel, reason } = this.fastChannel.detect(userInput);

    const context = await this.contextAssembler.assemble(characterId, userId, sessionId, userInput, {
      skipMemoryRetrieval: isFastChannel,
      seed: options.seed,
      now,
    });

    let text: string;
    let actions: string[];

    if (isFastChannel && reason) {
      const rule = await this.prisma.client.proceduralRule.findFirst({
        where: { characterId, situation: fastChannelSituation(reason) },
        orderBy: { weight: "desc" },
      });
      text = rule?.responsePattern ?? DEFAULT_FAST_REPLIES[reason];
      actions = extractActions(text);
    } else {
      const generated = await this.llmProvider.generate(context);
      text = generated.text;
      actions = generated.actions;
    }

    const forbiddenWords = getForbiddenWords(context.character.personalityArchetype);
    const filteredText = applySpeechQuirks(
      this.outputFilter.filter(text, forbiddenWords),
      context.character.personalityArchetype,
    );
    this.workingMemory.append(sessionId, { role: "character", content: filteredText, timestamp: now });

    return { text: filteredText, actions, isFastChannel, context };
  }
}