converse-mcp-server 2.22.8 → 2.25.1

package/src/tools/conversation.js

@@ -0,0 +1,1217 @@
+/**
+ * Conversation Tool
+ *
+ * Turn-based multi-model round-table. Models respond SEQUENTIALLY in the order
+ * given; each model sees the full running transcript (prior laps + earlier turns
+ * in the current lap) and builds on it. One tool call runs exactly one lap (one
+ * turn per model); the caller drives more laps by passing back the continuation_id.
+ *
+ * This is a sibling of consensus.js (parallel fan-out). It reuses the same
+ * infrastructure (context processing, model routing, custom-ID handling, async
+ * streaming, summarization, token limiting, export) but replaces the parallel
+ * two-phase core with a sequential lap loop.
+ *
+ * CRITICAL provider constraint: SDK providers (codex, claude, copilot) reduce the
+ * message array to ONLY the last `user` message. Therefore each turn's entire
+ * context (prior-lap transcript + lap prompt + same-lap turns + framing) is packed
+ * into a SINGLE self-contained final user message ("turn packet"). Do not spread
+ * turn context across multiple messages.
+ */
+
+import {
+  createToolResponse,
+  createToolError,
+  formatFailureDetails,
+} from './index.js';
+import {
+  createFileContext,
+} from '../utils/contextProcessor.js';
+import {
+  generateContinuationId,
+  isValidContinuationId,
+} from '../continuationStore.js';
+import { isSafeIdSegment } from '../utils/idValidation.js';
+import { debugLog, debugError } from '../utils/console.js';
+import { createLogger } from '../utils/logger.js';
+import { CONVERSATION_PROMPT } from '../systemPrompts.js';
+import { applyTokenLimit, getTokenLimit } from '../utils/tokenLimiter.js';
+import { validateAllPaths } from '../utils/fileValidator.js';
+import { SummarizationService } from '../services/summarizationService.js';
+import { exportConversation } from '../utils/conversationExporter.js';
+import {
+  mapModelToProvider,
+  resolveAutoModel,
+  getDefaultModelForProvider,
+} from '../utils/modelRouting.js';
+
+const logger = createLogger('conversation');
+
+/**
+ * Render the stored transcript (from prior laps) into labeled text that can be
+ * embedded in the next turn's packet. Stored state pairs user (lap prompt) and
+ * assistant (lap transcript) messages; we re-render those as readable context so
+ * last-user-only SDK providers still see the history (and so a provider does not
+ * mistake prior multi-speaker transcript for its own previous output).
+ * @param {Array} storedMessages - Stored messages from a prior conversation state
+ * @returns {string} Labeled prior-transcript text ('' for a new conversation)
+ */
+function renderStoredTranscriptToText(storedMessages = []) {
+  if (!Array.isArray(storedMessages) || storedMessages.length === 0) {
+    return '';
+  }
+
+  const blocks = [];
+  let lapNumber = 0;
+  let pendingPrompt = null;
+
+  const toText = (content) => {
+    if (typeof content === 'string') {
+      return content;
+    }
+    if (Array.isArray(content)) {
+      // Complex content array (files/images + text) — extract text parts only
+      return content
+        .filter((part) => part && part.type === 'text' && part.text)
+        .map((part) => part.text)
+        .join('\n');
+    }
+    return '';
+  };
+
+  for (const message of storedMessages) {
+    if (!message || message.role === 'system') {
+      continue;
+    }
+    if (message.role === 'user') {
+      pendingPrompt = toText(message.content);
+    } else if (message.role === 'assistant') {
+      lapNumber += 1;
+      const promptText = pendingPrompt ? `${pendingPrompt}\n\n` : '';
+      const assistantText = toText(message.content);
+      blocks.push(
+        `## Earlier in this round-table (lap ${lapNumber}):\n${promptText}${assistantText}`,
+      );
+      pendingPrompt = null;
+    }
+  }
+
+  return blocks.join('\n\n');
+}
+
+/**
+ * Build the per-turn framing text for the model at position `i`.
+ * @param {object} params
+ * @returns {string} Framing text appended to the turn packet
+ */
+function buildFramingText({ i, models, turn_prompt }) {
+  const total = models.length;
+  const selfModel = models[i];
+  const prevModel = i > 0 ? models[i - 1] : null;
+  const nextModel = i < total - 1 ? models[i + 1] : null;
+
+  const order = models.join(', ');
+  const prevText = prevModel || 'no one (you open the round)';
+  const nextText = nextModel || 'no one (you close this round)';
+  const handoffText = nextModel
+    ? `Your response will be passed to the next participant (${nextModel}).`
+    : 'Your response will be returned to the user, as you are the last participant this round.';
+
+  const lines = [
+    `You are participant "${selfModel}" in a multi-model round-table conversation.`,
+    `Participants, in speaking order: ${order}.`,
+    `You are speaking in position ${i + 1} of ${total}, after ${prevText}, before ${nextText}.`,
+    'The original topic/prompt for this round is shown above, followed by any responses already given this round.',
+    'Respond to the whole conversation so far — build on, challenge, or refine what others have said; do not merely repeat them.',
+    handoffText,
+  ];
+
+  if (turn_prompt && typeof turn_prompt === 'string' && turn_prompt.trim()) {
+    lines.push(turn_prompt.trim());
+  }
+
+  return lines.join('\n');
+}
+
+/**
+ * Build the single self-contained turn packet TEXT for the model at position `i`.
+ * Order: prior-transcript section, lap prompt, same-lap turns, framing.
+ * This is the LAST user message — the only thing last-user-only SDK providers see.
+ * @param {object} params
+ * @returns {string} Turn packet text
+ */
+function buildTurnPacket({
+  priorTranscriptText,
+  prompt,
+  sameLapTurns,
+  i,
+  models,
+  turn_prompt,
+}) {
+  const parts = [];
+
+  if (priorTranscriptText && priorTranscriptText.trim()) {
+    parts.push(priorTranscriptText.trim());
+  }
+
+  parts.push(`Original topic for this round:\n${prompt}`);
+
+  // Same-lap turns from models 0..i-1 (omitted for the opener, i=0)
+  if (i > 0 && sameLapTurns.length > 0) {
+    const turnBlocks = sameLapTurns.map((turn) => {
+      if (turn.status === 'success') {
+        return `### ${turn.model} said:\n${turn.response}`;
+      }
+      return `### ${turn.model} did not respond (error: ${turn.error})`;
+    });
+    parts.push(turnBlocks.join('\n\n'));
+  }
+
+  parts.push(buildFramingText({ i, models, turn_prompt }));
+
+  return parts.join('\n\n');
+}
+
+/**
+ * Format the full lap transcript for storage/display.
+ * @param {Array} lapTurns - Turns from the current lap
+ * @returns {string} Formatted transcript
+ */
+function formatLapTranscript(lapTurns) {
+  let content = '';
+  let successful = 0;
+
+  lapTurns.forEach((turn, index) => {
+    if (turn.status === 'success') {
+      successful += 1;
+      content += `### ${turn.model} (turn ${index + 1}):\n${turn.response}\n\n---\n\n`;
+    } else {
+      content += `### ${turn.model} (turn ${index + 1}, did not respond):\nError: ${turn.error}\n\n---\n\n`;
+    }
+  });
+
+  content += `\n**Summary:** Conversation lap completed with ${successful}/${lapTurns.length} successful turns.`;
+  return content;
+}
+
+/**
+ * Resolve the ordered model list into a turn plan. Unlike consensus, unknown or
+ * unavailable models are NOT dropped — they are recorded with a preFailReason so
+ * they keep their position in the order (and produce a failed turn).
+ * @param {Array<string>} models - Ordered model list
+ * @param {object} providers - Provider instances
+ * @param {object} config - Configuration
+ * @returns {Array<object>} Ordered turn plan entries
+ */
+function resolveTurnPlan(models, providers, config) {
+  // Single "auto" expands to the first available provider's default model only
+  // (a single-model round-table is valid). Multiple explicit models resolve per-entry.
+  let modelsToProcess = models;
+  if (models.length === 1 && String(models[0]).toLowerCase() === 'auto') {
+    const providerOrder = [
+      'codex',
+      'gemini-cli',
+      'claude',
+      'copilot',
+      'openai',
+      'google',
+      'xai',
+      'anthropic',
+      'mistral',
+      'deepseek',
+      'openrouter',
+    ];
+
+    let firstAvailable = null;
+    for (const providerName of providerOrder) {
+      const provider = providers[providerName];
+      if (provider && provider.isAvailable(config)) {
+        firstAvailable = providerName;
+        break;
+      }
+    }
+
+    // If a provider is available, use its default model. Otherwise keep "auto"
+    // so it resolves to a turn that fails cleanly (all-fail laps must complete).
+    modelsToProcess = firstAvailable
+      ? [getDefaultModelForProvider(firstAvailable)]
+      : ['auto'];
+  }
+
+  return modelsToProcess.map((modelName) => {
+    if (!modelName || typeof modelName !== 'string') {
+      return {
+        model: modelName || 'unknown',
+        provider: null,
+        providerInstance: null,
+        resolvedModel: null,
+        preFailReason: 'Invalid model specification',
+      };
+    }
+
+    const providerName = mapModelToProvider(modelName, providers);
+    const resolvedModel = resolveAutoModel(modelName, providerName);
+    const provider = providers[providerName];
+
+    if (!provider) {
+      return {
+        model: modelName,
+        provider: providerName,
+        providerInstance: null,
+        resolvedModel,
+        preFailReason: `Provider not found: ${providerName}`,
+      };
+    }
+
+    if (!provider.isAvailable(config)) {
+      return {
+        model: modelName,
+        provider: providerName,
+        providerInstance: null,
+        resolvedModel,
+        preFailReason: `Provider ${providerName} not available (check API key)`,
+      };
+    }
+
+    return {
+      model: modelName,
+      provider: providerName,
+      providerInstance: provider,
+      resolvedModel,
+      preFailReason: null,
+    };
+  });
+}
+
+/**
+ * Process files/images into a context message (shared sync + async helper).
+ * @returns {Promise<object|null>} Context message or null
+ */
+async function buildContextMessage(files, images, contextProcessor, config) {
+  if (files.length === 0 && images.length === 0) {
+    return null;
+  }
+
+  try {
+    const contextRequest = {
+      files: Array.isArray(files) ? files : [],
+      images: Array.isArray(images) ? images : [],
+    };
+
+    const contextResult = await contextProcessor.processUnifiedContext(
+      contextRequest,
+      {
+        enforceSecurityCheck: false,
+        skipSecurityCheck: true,
+        clientCwd: config.server?.client_cwd,
+      },
+    );
+
+    const allProcessedFiles = [
+      ...contextResult.files,
+      ...contextResult.images,
+    ];
+    if (allProcessedFiles.length > 0) {
+      return createFileContext(allProcessedFiles, {
+        includeMetadata: true,
+        includeErrors: true,
+      });
+    }
+  } catch (error) {
+    logger.error('Error processing context', { error });
+    // Continue without context if processing fails
+  }
+
+  return null;
+}
+
+/**
+ * Build the final user message content for a turn. Files/images are attached to
+ * THIS message so multimodal providers see them, with the packet text appended.
+ * @returns {string|Array} User message content
+ */
+function buildTurnUserContent(packetText, contextMessage) {
+  if (contextMessage && contextMessage.content) {
+    return [...contextMessage.content, { type: 'text', text: packetText }];
+  }
+  return packetText;
+}
+
+/**
+ * Build the persisted conversation state for a completed lap. Mirrors consensus's
+ * `[...messages, assistantMessage]` shape: one system message at index 0 (added
+ * for a fresh conversation), accumulating user (lap prompt) / assistant (lap
+ * transcript) pairs.
+ * @returns {object} Conversation state to persist
+ */
+function buildConversationState(
+  priorMessages,
+  lapUserMessage,
+  assistantMessage,
+  models,
+  turnsSuccessful,
+  turnsFailed,
+) {
+  // priorMessages is the loaded stored history (may include a leading system msg).
+  const hasSystem =
+    priorMessages.length > 0 && priorMessages[0].role === 'system';
+
+  const baseMessages = hasSystem
+    ? priorMessages
+    : [{ role: 'system', content: CONVERSATION_PROMPT }, ...priorMessages];
+
+  return {
+    messages: [...baseMessages, lapUserMessage, assistantMessage],
+    type: 'conversation',
+    lastUpdated: Date.now(),
+    conversationData: {
+      modelsOrdered: models,
+      turnsSuccessful,
+      turnsFailed,
+    },
+  };
+}
+
+/**
+ * Conversation tool implementation
+ * @param {object} args - Tool arguments
+ * @param {object} dependencies - Injected dependencies
+ * @returns {object} MCP tool response
+ */
+export async function conversationTool(args, dependencies) {
+  try {
+    const {
+      config,
+      providers,
+      continuationStore,
+      contextProcessor,
+      jobRunner,
+      providerStreamNormalizer,
+      signal,
+    } = dependencies;
+
+    // Validate required arguments
+    if (
+      !args.prompt ||
+      typeof args.prompt !== 'string' ||
+      !args.prompt.trim()
+    ) {
+      return createToolError('Prompt is required and must be a string');
+    }
+
+    if (
+      !args.models ||
+      !Array.isArray(args.models) ||
+      args.models.length === 0
+    ) {
+      return createToolError(
+        'Models array is required and must contain at least one model',
+      );
+    }
+
+    // Extract and validate arguments
+    const {
+      prompt,
+      models,
+      files = [],
+      images = [],
+      continuation_id,
+      temperature = 0.2,
+      reasoning_effort = 'medium',
+      use_websearch = false,
+      async = false,
+      export: shouldExport = false,
+      turn_prompt,
+    } = args;
+
+    // Handle async execution mode
+    if (async) {
+      if (!jobRunner || !providerStreamNormalizer) {
+        return createToolError(
+          'Async execution not available - missing async dependencies',
+        );
+      }
+
+      // Validate custom continuation ID for async safety (used as path segment)
+      if (continuation_id && !isSafeIdSegment(continuation_id)) {
+        return createToolError(
+          `Invalid continuation_id for async mode: "${continuation_id}". Async IDs must contain only letters, numbers, hyphens, and underscores (max 128 chars).`,
+        );
+      }
+
+      const bgContinuationId = continuation_id || generateContinuationId();
+
+      // Determine if this is a custom ID (non-standard format AND not found in store)
+      let isCustomId = false;
+      if (continuation_id && !isValidContinuationId(continuation_id)) {
+        try {
+          const existing = await continuationStore.get(continuation_id);
+          isCustomId = !existing;
+        } catch {
+          isCustomId = true;
+        }
+      }
+
+      const modelsList = args.models.join(', ');
+
+      // Generate title early for initial response
+      const summarizationService = new SummarizationService(providers, config);
+      let title = null;
+      try {
+        title = await summarizationService.generateTitle(prompt);
+        debugLog(
+          `Conversation: Generated title for initial response - "${title}"`,
+        );
+      } catch (error) {
+        debugError(
+          'Conversation: Failed to generate title for initial response',
+          error,
+        );
+        title = prompt.substring(0, 50);
+      }
+
+      try {
+        await jobRunner.submit(
+          {
+            tool: 'conversation',
+            sessionId: bgContinuationId,
+            options: {
+              ...args,
+              jobId: bgContinuationId,
+              models_list: modelsList,
+              title,
+            },
+          },
+          async (context) => {
+            return await executeConversationWithStreaming(
+              args,
+              {
+                ...dependencies,
+                continuationId: bgContinuationId,
+                isCustomId,
+                title,
+              },
+              context,
+            );
+          },
+        );
+
+        const startTime = new Date()
+          .toLocaleString('en-GB', {
+            day: '2-digit',
+            month: '2-digit',
+            year: 'numeric',
+            hour: '2-digit',
+            minute: '2-digit',
+            second: '2-digit',
+            hour12: false,
+          })
+          .replace(',', '');
+
+        const statusLine = `⏳ SUBMITTED | CONVERSATION | ${bgContinuationId} | 1/1 | Started: ${startTime} | "${title || 'Processing...'}" | ${modelsList}`;
+
+        return createToolResponse({
+          content: `${statusLine}\ncontinuation_id: ${bgContinuationId}`,
+          continuation: {
+            id: bgContinuationId,
+            status: 'processing',
+            ...(isCustomId && { custom_id: true }),
+          },
+          async_execution: true,
+        });
+      } catch (error) {
+        logger.error('Failed to submit async conversation job', { error });
+        return createToolError(`Async execution failed: ${error.message}`);
+      }
+    }
+
+    // --- Synchronous path ---
+
+    let conversationHistory = [];
+    let continuationId = continuation_id;
+    let isCustomId = false;
+
+    // Load existing conversation if continuation_id provided
+    if (continuationId) {
+      try {
+        const existingState = await continuationStore.get(continuationId);
+        if (existingState) {
+          conversationHistory = existingState.messages || [];
+        } else {
+          // Preserve user-provided ID and start fresh conversation
+          isCustomId = !isValidContinuationId(continuationId);
+        }
+      } catch (error) {
+        logger.error('Error loading conversation', { error });
+        isCustomId = !isValidContinuationId(continuationId);
+      }
+    } else {
+      continuationId = generateContinuationId();
+    }
+
+    // Validate file paths before processing
+    if (files.length > 0 || images.length > 0) {
+      const validation = await validateAllPaths(
+        { files, images },
+        { clientCwd: config.server?.client_cwd },
+      );
+      if (!validation.valid) {
+        logger.error('File validation failed', { errors: validation.errors });
+        return validation.errorResponse;
+      }
+    }
+
+    const contextMessage = await buildContextMessage(
+      files,
+      images,
+      contextProcessor,
+      config,
+    );
+
+    // Re-render prior stored laps into labeled text for the turn packets
+    const priorTranscriptText = renderStoredTranscriptToText(
+      conversationHistory,
+    );
+
+    // Resolve ordered turn plan (unavailable models kept as pre-failed turns)
+    const turnPlan = resolveTurnPlan(models, providers, config);
+
+    const startedAt = Date.now();
+    const lapTurns = [];
+
+    // Sequential lap loop: one turn per model, in order
+    for (let i = 0; i < turnPlan.length; i++) {
+      // Honor cancellation between turns
+      if (signal?.aborted) {
+        logger.debug('Conversation tool cancelled by client mid-lap');
+        return createToolError('Conversation request cancelled');
+      }
+
+      const plan = turnPlan[i];
+
+      if (plan.preFailReason) {
+        lapTurns.push({
+          model: plan.model,
+          provider: plan.provider,
+          status: 'failed',
+          error: plan.preFailReason,
+          position: i,
+        });
+        continue;
+      }
+
+      const packetText = buildTurnPacket({
+        priorTranscriptText,
+        prompt,
+        sameLapTurns: lapTurns,
+        i,
+        models,
+        turn_prompt,
+      });
+
+      const finalUserContent = buildTurnUserContent(packetText, contextMessage);
+
+      const messages = [
+        { role: 'system', content: CONVERSATION_PROMPT },
+        { role: 'user', content: finalUserContent },
+      ];
+
+      try {
+        const response = await plan.providerInstance.invoke(messages, {
+          temperature,
+          reasoning_effort,
+          use_websearch,
+          signal,
+          config,
+          model: plan.resolvedModel,
+        });
+
+        lapTurns.push({
+          model: plan.model,
+          provider: plan.provider,
+          status: 'success',
+          response: response.content,
+          metadata: response.metadata || {},
+          position: i,
+        });
+      } catch (error) {
+        if (signal?.aborted || error.name === 'AbortError') {
+          logger.debug('Conversation tool cancelled during turn');
+          return createToolError('Conversation request cancelled');
+        }
+        lapTurns.push({
+          model: plan.model,
+          provider: plan.provider,
+          status: 'failed',
+          error: error.message,
+          position: i,
+        });
+      }
+    }
+
+    const turnsSuccessful = lapTurns.filter(
+      (t) => t.status === 'success',
+    ).length;
+    const turnsFailed = lapTurns.length - turnsSuccessful;
+
+    // Build the lap user message (lap prompt, with context if present)
+    const lapUserMessage = {
+      role: 'user',
+      content: buildTurnUserContent(prompt, contextMessage),
+    };
+
+    // Labeled lap transcript (### <model> (turn <n>):) — computed once and reused
+    // for the assistant message, the persisted state, and the response content.
+    const transcript = formatLapTranscript(lapTurns);
+
+    const assistantMessage = {
+      role: 'assistant',
+      content: transcript,
+    };
+
+    // Save conversation state (skip on abort to avoid persisting incomplete history)
+    let conversationState;
+    if (!signal?.aborted) {
+      try {
+        conversationState = buildConversationState(
+          conversationHistory,
+          lapUserMessage,
+          assistantMessage,
+          models,
+          turnsSuccessful,
+          turnsFailed,
+        );
+
+        await continuationStore.set(continuationId, conversationState);
+      } catch (error) {
+        logger.error('Error saving conversation', { error });
+        // Continue even if save fails
+      }
+    }
+
+    // Export conversation if requested
+    if (shouldExport && conversationState) {
+      await exportConversation(conversationState, {
+        clientCwd: config.server?.client_cwd,
+        continuation_id: continuationId,
+        models,
+        temperature,
+        reasoning_effort,
+        use_websearch,
+        files,
+        images,
+      });
+    }
+
+    const executionTime = (Date.now() - startedAt) / 1000;
+    const messageCount = (conversationState?.messages || []).length;
+
+    // Collect failure details
+    const failureDetails = lapTurns
+      .filter((t) => t.status === 'failed')
+      .map((t) => `${t.model} (${t.error})`);
+
+    const modelsList = models.join(', ');
+    const statusLine =
+      config.environment?.nodeEnv !== 'test'
+        ? `✅ COMPLETED | CONVERSATION | ${continuationId} | ${executionTime.toFixed(1)}s elapsed | ${turnsSuccessful}/${lapTurns.length} turns | ${modelsList}\n`
+        : '';
+
+    const continuationIdLine = `continuation_id: ${continuationId}\n\n`;
+
+    const result = {
+      status: 'conversation_complete',
+      content: transcript,
+      models_consulted: models.length,
+      successful_turns: turnsSuccessful,
+      failed_turns: turnsFailed,
+      turns: lapTurns,
+      continuation: {
+        id: continuationId,
+        messageCount,
+        ...(isCustomId && { custom_id: true }),
+      },
+      settings: {
+        temperature,
+        models_requested: models,
+      },
+    };
+
+    const tokenLimit = getTokenLimit(config);
+    const resultStr = JSON.stringify(result, null, 2);
+    const limitedResult = applyTokenLimit(resultStr, tokenLimit);
+
+    let finalContent = limitedResult.content;
+    if (failureDetails.length > 0) {
+      finalContent += formatFailureDetails(failureDetails);
+    }
+
+    finalContent = statusLine + continuationIdLine + finalContent;
+
+    return createToolResponse({
+      content: finalContent,
+      continuation: {
+        id: continuationId,
+        messageCount,
+        ...(isCustomId && { custom_id: true }),
+      },
+    });
+  } catch (error) {
+    if (dependencies?.signal?.aborted || error.name === 'AbortError') {
+      logger.debug('Conversation tool cancelled by client');
+      return createToolError('Conversation request cancelled');
+    }
+    logger.error('Conversation tool error', { error });
+    return createToolError('Conversation tool failed', error);
+  }
+}
+
+/**
+ * Execute a single turn with streaming support (async path). Adapts consensus's
+ * per-provider streaming to a single provider per call.
+ * @returns {Promise<object>} Turn result { model, provider, status, response|error }
+ */
+async function executeTurnWithStreaming(
+  plan,
+  messages,
+  options,
+  context,
+  streamNormalizer,
+  turnIndex,
+) {
+  try {
+    if (context.signal?.aborted) {
+      throw new Error('Conversation execution was cancelled');
+    }
+
+    let response;
+    let stream = null;
+
+    if (
+      plan.providerInstance.stream &&
+      typeof plan.providerInstance.stream === 'function'
+    ) {
+      stream = plan.providerInstance.stream(messages, options);
+    } else {
+      // SDK providers (copilot, codex, claude, gemini-cli) stream via invoke
+      const streamResult = await plan.providerInstance.invoke(messages, {
+        ...options,
+        stream: true,
+      });
+      if (
+        streamResult &&
+        typeof streamResult[Symbol.asyncIterator] === 'function'
+      ) {
+        stream = streamResult;
+      } else {
+        response = streamResult;
+      }
+    }
+
+    if (stream) {
+      const normalizedStream = streamNormalizer.normalize(plan.provider, stream, {
+        provider: plan.provider,
+        model: options.model,
+        requestId: `${context.jobId}-turn-${turnIndex}`,
+      });
+
+      let accumulatedContent = '';
+      let finalUsage = null;
+      let finalMetadata = {};
+
+      for await (const event of normalizedStream) {
+        if (context.signal?.aborted) {
+          throw new Error('Conversation execution was cancelled');
+        }
+
+        switch (event.type) {
+        case 'delta':
+          accumulatedContent += event.data.textDelta;
+          break;
+        case 'usage':
+          finalUsage = event.data.usage;
+          break;
+        case 'end':
+          accumulatedContent = event.data.content || accumulatedContent;
+          finalUsage = event.data.usage || finalUsage;
+          finalMetadata = event.data.metadata || finalMetadata;
+          break;
+        case 'error':
+          throw new Error(`Streaming error: ${event.data.error.message}`);
+        }
+      }
+
+      response = {
+        content: accumulatedContent,
+        metadata: { ...finalMetadata, usage: finalUsage, streaming: true },
+      };
+    }
+
+    if (!stream && !response) {
+      response = await plan.providerInstance.invoke(messages, options);
+    }
+
+    return {
+      model: plan.model,
+      provider: plan.provider,
+      status: 'success',
+      response: response.content,
+      metadata: response.metadata || {},
+    };
+  } catch (error) {
+    // Cancellation must abort the whole lap, not be demoted to a failed turn.
+    // Rethrow so it propagates out of executeConversationWithStreaming to the
+    // job runner (which marks the job cancelled) and the save block is skipped.
+    if (context.signal?.aborted || error.name === 'AbortError') {
+      throw error;
+    }
+    return {
+      model: plan.model,
+      provider: plan.provider,
+      status: 'failed',
+      error: error.message,
+    };
+  }
+}
+
+/**
+ * Execute a conversation lap with streaming normalization for async execution.
+ * Mirrors executeConsensusWithStreaming but sequential.
+ * @param {object} args - Original conversation arguments
+ * @param {object} dependencies - Dependencies with continuationId
+ * @param {object} context - Job execution context
+ * @returns {Promise<object>} Complete conversation result (with top-level content)
+ */
+async function executeConversationWithStreaming(args, dependencies, context) {
+  const {
+    config,
+    providers,
+    continuationStore,
+    contextProcessor,
+    providerStreamNormalizer,
+    continuationId,
+    isCustomId,
+    title: passedTitle,
+  } = dependencies;
+
+  const {
+    prompt,
+    models,
+    files = [],
+    images = [],
+    temperature = 0.2,
+    reasoning_effort = 'medium',
+    use_websearch = false,
+    export: shouldExport = false,
+    turn_prompt,
+  } = args;
+
+  let conversationHistory = [];
+  if (continuationId) {
+    try {
+      const existingState = await continuationStore.get(continuationId);
+      if (existingState) {
+        conversationHistory = existingState.messages || [];
+      }
+    } catch (error) {
+      logger.error('Error loading conversation', { error });
+    }
+  }
+
+  // Validate file paths before processing
+  if (files.length > 0 || images.length > 0) {
+    const validation = await validateAllPaths(
+      { files, images },
+      { clientCwd: config.server?.client_cwd },
+    );
+    if (!validation.valid) {
+      logger.error('File validation failed', { errors: validation.errors });
+      throw new Error(
+        `File validation failed: ${validation.errors.join(', ')}`,
+      );
+    }
+  }
+
+  const contextMessage = await buildContextMessage(
+    files,
+    images,
+    contextProcessor,
+    config,
+  );
+
+  const priorTranscriptText = renderStoredTranscriptToText(conversationHistory);
+  const turnPlan = resolveTurnPlan(models, providers, config);
+  const modelsList = models.join(', ');
+
+  // Use passed title or generate if not provided
+  const summarizationService = new SummarizationService(providers, config);
+  let title = passedTitle;
+  if (!title) {
+    try {
+      title = await summarizationService.generateTitle(prompt);
+      debugLog(`Conversation: Generated title - "${title}"`);
+    } catch (error) {
+      debugError('Conversation: Error generating title', error);
+      title = prompt.substring(0, 50);
+    }
+  }
+
+  await context.updateJob({
+    models_list: modelsList,
+    title,
+    conversation_progress: `0/${turnPlan.length}`,
+    conversation_phase: 'conversation',
+    total_turns: turnPlan.length,
+    completed_turns: 0,
+  });
+
+  const startedAt = Date.now();
+  const lapTurns = [];
+
+  for (let i = 0; i < turnPlan.length; i++) {
+    if (context.signal?.aborted) {
+      throw new Error('Conversation execution was cancelled');
+    }
+
+    const plan = turnPlan[i];
+
+    if (plan.preFailReason) {
+      lapTurns.push({
+        model: plan.model,
+        provider: plan.provider,
+        status: 'failed',
+        error: plan.preFailReason,
+        position: i,
+      });
+    } else {
+      const packetText = buildTurnPacket({
+        priorTranscriptText,
+        prompt,
+        sameLapTurns: lapTurns,
+        i,
+        models,
+        turn_prompt,
+      });
+
+      const finalUserContent = buildTurnUserContent(packetText, contextMessage);
+
+      const messages = [
+        { role: 'system', content: CONVERSATION_PROMPT },
+        { role: 'user', content: finalUserContent },
+      ];
+
+      const turnResult = await executeTurnWithStreaming(
+        plan,
+        messages,
+        {
+          temperature,
+          reasoning_effort,
+          use_websearch,
+          signal: context?.signal,
+          config,
+          model: plan.resolvedModel,
+        },
+        context,
+        providerStreamNormalizer,
+        i,
+      );
+
+      lapTurns.push({ ...turnResult, position: i });
+    }
+
+    // Report per-turn progress with the running transcript.
+    // Use flat keys (not a `progress` object) — asyncJobStore.update() treats the
+    // reserved `progress` key as a numeric 0..1 value, so an object there would
+    // corrupt it. Numeric overall progress is supplied separately as a fraction.
+    await context.updateJob({
+      conversation_progress: `${i + 1}/${turnPlan.length}`,
+      accumulated_content: formatLapTranscript(lapTurns),
+      title,
+      progress: (i + 1) / turnPlan.length,
+      conversation_phase: 'conversation',
+      total_turns: turnPlan.length,
+      completed_turns: i + 1,
+      current_model: plan.model,
+    });
+  }
+
+  const turnsSuccessful = lapTurns.filter((t) => t.status === 'success').length;
+  const turnsFailed = lapTurns.length - turnsSuccessful;
+
+  const lapUserMessage = {
+    role: 'user',
+    content: buildTurnUserContent(prompt, contextMessage),
+  };
+
+  // Final lap transcript — computed once and reused for the assistant message,
+  // the persisted state, and the returned top-level content.
+  const transcript = formatLapTranscript(lapTurns);
+
+  const assistantMessage = {
+    role: 'assistant',
+    content: transcript,
+  };
+
+  // Save conversation state
+  let conversationState;
+  try {
+    conversationState = buildConversationState(
+      conversationHistory,
+      lapUserMessage,
+      assistantMessage,
+      models,
+      turnsSuccessful,
+      turnsFailed,
+    );
+    await continuationStore.set(continuationId, conversationState);
+  } catch (error) {
+    logger.error('Error saving conversation', { error });
+  }
+
+  // Export conversation if requested
+  if (shouldExport && conversationState) {
+    await exportConversation(conversationState, {
+      clientCwd: config.server?.client_cwd,
+      continuation_id: continuationId,
+      models,
+      temperature,
+      reasoning_effort,
+      use_websearch,
+      files,
+      images,
+    });
+  }
+
+  const executionTime = (Date.now() - startedAt) / 1000;
+
+  // Generate final summary from combined successful responses
+  let finalSummary = null;
+  const combinedResponses = lapTurns
+    .filter((t) => t.status === 'success' && t.response)
+    .map((t) => `${t.model}:\n${t.response}`);
+
+  if (combinedResponses.length > 0) {
+    const combinedContent = combinedResponses.join('\n\n---\n\n');
+    if (combinedContent.length > 100) {
+      try {
+        finalSummary =
+          await summarizationService.generateFinalSummary(combinedContent);
+        debugLog(`Conversation: Generated final summary - "${finalSummary}"`);
+        await context.updateJob({ final_summary: finalSummary });
+      } catch (error) {
+        debugError('Conversation: Error generating final summary', error);
+      }
+    }
+  }
+
+  const failureDetails = lapTurns
+    .filter((t) => t.status === 'failed')
+    .map((t) => `${t.model} (${t.error})`);
+
+  const messageCount = (conversationState?.messages || []).length;
+
+  // Top-level `content` is required: formatStatus only renders result.content
+  // when displaying a completed async job.
+  return {
+    status: 'conversation_complete',
+    content: transcript,
+    models_consulted: models.length,
+    successful_turns: turnsSuccessful,
+    failed_turns: turnsFailed,
+    turns: lapTurns,
+    continuation: {
+      id: continuationId,
+      messageCount,
+      ...(isCustomId && { custom_id: true }),
+    },
+    settings: {
+      temperature,
+      models_requested: models,
+    },
+    metadata: {
+      execution_time: executionTime,
+      async_execution: true,
+      successful_models: turnsSuccessful,
+      total_models: models.length,
+      failure_details: failureDetails,
+      title,
+      final_summary: finalSummary,
+    },
+  };
+}
+
+// Tool metadata
+conversationTool.description =
+  'TURN-BASED ROUND-TABLE - Models respond SEQUENTIALLY in the order given; each model sees the full running transcript and builds on prior turns. One call = one lap; pass continuation_id for more laps. Contrast with consensus (parallel, same prompt). Use the "files" parameter to share code.';
+conversationTool.inputSchema = {
+  type: 'object',
+  properties: {
+    models: {
+      type: 'array',
+      items: { type: 'string' },
+      minItems: 1,
+      description:
+        'Ordered list of models for the round-table. ORDER MATTERS: models speak one after another in this exact order, each seeing the transcript of those before it. Examples: ["codex", "gemini", "claude"]. A single model (e.g. ["codex"]) talks to itself across laps. Use ["auto"] to pick the first available provider.',
+    },
+    prompt: {
+      type: 'string',
+      description:
+        'The topic or question to open the round-table with. Include context and what you want the participants to discuss. Example: "Critique this caching strategy and propose improvements."',
+    },
+    continuation_id: {
+      type: 'string',
+      description:
+        'Thread continuation ID for running more laps. Auto-generated in the first response; pass it back to run another lap where every model again sees the full accumulated transcript. You MAY change the models list on a resuming lap.',
+    },
+    turn_prompt: {
+      type: 'string',
+      description:
+        'Optional custom per-turn instruction appended to the round-table framing each model receives. Example: "Focus on security implications in your turn."',
+    },
+    files: {
+      type: 'array',
+      items: { type: 'string' },
+      description:
+        'File paths for additional context (absolute or relative paths). Supports line ranges: file.txt{10:50}, file.txt{100:}. Files are shared with every participant in the lap. IMPORTANT: Always use this parameter to share file content instead of copying code into the prompt.',
+    },
+    images: {
+      type: 'array',
+      items: { type: 'string' },
+      description:
+        'Image paths for visual context (absolute or relative paths, or base64). Example: ["C:\\Users\\username\\diagram.png", "./flow.jpg"]',
+    },
+    temperature: {
+      type: 'number',
+      description:
+        'Response randomness (0.0-1.0). Examples: 0.1 (very focused), 0.2 (analytical - default), 0.5 (balanced). Default: 0.2',
+      minimum: 0.0,
+      maximum: 1.0,
+      default: 0.2,
+    },
+    reasoning_effort: {
+      type: 'string',
+      enum: ['none', 'minimal', 'low', 'medium', 'high', 'max'],
+      description:
+        'Reasoning depth for thinking models. Examples: "none" (no reasoning, fastest), "low" (light analysis), "medium" (balanced), "high" (complex analysis). Default: "medium"',
+      default: 'medium',
+    },
+    use_websearch: {
+      type: 'boolean',
+      description:
+        'Enable web search for current information. Only works with models that support web search (OpenAI, XAI, Google). Default: false',
+      default: false,
+    },
+    async: {
+      type: 'boolean',
+      description:
+        'Execute the lap in background with per-turn progress tracking. When true, returns continuation_id immediately and processes the lap asynchronously. Default: false',
+      default: false,
+    },
+    export: {
+      type: 'boolean',
+      description:
+        'Export conversation to disk. Creates folder with continuation_id name containing numbered request/response files and metadata. Default: false',
+      default: false,
+    },
+  },
+  required: ['prompt', 'models'],
+};