@flamingo-stack/openframe-frontend-core 0.0.467 → 0.0.468-snapshot.20260723140929

package/src/components/chat/utils/process-historical-messages.ts

@@ -1,6 +1,28 @@
 /**
- * Utility for processing historical messages from GraphQL/API
- * into display-ready format
+ * Utility for processing historical messages from GraphQL/API into
+ * display-ready format.
+ *
+ * Phase 3 of the chat unification reduced this file to DECODE + ENVELOPE:
+ *
+ *   - `decodeHistoricalMessageData` maps one persisted `MessageData` item
+ *     into the shared `ChatStreamEvent` vocabulary (historical items share
+ *     the MESSAGE_TYPE vocabulary with NATS chunks but are NOT full chunk
+ *     envelopes — e.g. compaction summaries ride in `summary`, not `text`,
+ *     and APPROVAL_RESULT rows carry `resolvedByName` directly);
+ *   - `applyHistoryEvent` replays each decoded event into the shared
+ *     per-turn kernel (`MessageSegmentAccumulator` — the same kernel the
+ *     master `createChatStreamReducer` instantiates) with the
+ *     history-specific approval semantics (display-all default, pending
+ *     tracking + `flushPendingApprovals`, `approvalStatuses` overrides);
+ *   - the ENVELOPE (user/assistant flush-grouping, OWNER_TYPE author /
+ *     display-name / avatar resolution, standalone SYSTEM handling,
+ *     per-message contextItems mapping, per-turn streamSeq MAX) lives ONCE
+ *     in `processHistory` — the previous near-duplicate second copy was
+ *     deleted (`processHistoricalMessagesWithErrors` is an alias; the two
+ *     snapshots were byte-identical).
+ *
+ * Adapters then feed the processed rows into the reducer via
+ * `initializeWithState` (see `useNatsChatAdapter.loadDialogHistory`).
  */
 
 import {
@@ -15,6 +37,7 @@ import {
   type MessageData,
   type MessageOwner,
 } from '../types'
+import type { ChatStreamEvent } from '../../../chat-protocol/events'
 import { MessageSegmentAccumulator, createMessageSegmentAccumulator } from './message-segment-accumulator'
 import { getCommandText } from './tool-call-helpers'
 
@@ -53,232 +76,45 @@ function pushStandaloneMessages(
   })
 }
 
-/**
- * Result type for historical message processing
- */
-export interface ProcessHistoricalMessagesResult {
-  messages: ProcessedMessage[]
-  escalatedApprovals: Map<
-    string,
-    { command: string; explanation?: string; approvalType: string; toolCalls?: PendingToolCallData[] }
-  >
-}
+// =============================================================================
+// Decode — persisted MessageData item → shared ChatStreamEvent vocabulary
+// =============================================================================
 
 /**
- * Process an array of historical messages into display-ready format
+ * Map one persisted `MessageData` item to a normalized `ChatStreamEvent`.
+ * Thin shim over the NATS chunk decoder's core mapping — the differences
+ * are exactly the persisted-row divergences from the realtime envelope:
+ *
+ *   - compaction summaries arrive in `summary` (realtime: `text`);
+ *   - APPROVAL_RESULT rows carry `resolvedByName` (realtime chunks carry
+ *     the resolver's name as `displayName`);
+ *   - `approvalType` defaults to 'CLIENT' (legacy history parity;
+ *     realtime defaults APPROVAL_REQUEST to 'USER');
+ *   - batch `toolCalls` pass through verbatim (already the persisted
+ *     `PendingToolCallData` shape — no normalization pass).
+ *
+ * Returns `null` for rows the assistant-turn path doesn't decode (SYSTEM
+ * is handled by the envelope's standalone path; user TEXT rows are handled
+ * by the envelope's user path).
  */
-export function processHistoricalMessages(
-  messages: HistoricalMessage[],
-  options: MessageProcessingOptions = {}
-): ProcessHistoricalMessagesResult {
-  const {
-    assistantName = 'Fae',
-    assistantType = 'fae',
-    assistantAvatar,
-    onApprove,
-    onReject,
-    chatTypeFilter,
-    approvalStatuses = {},
-    // An omitted option means "display every approval type" — the original
-    // history semantics. Deliberately NOT defaulted to the realtime
-    // processor's ['CLIENT']: consumers that pass a wider list to their
-    // realtime processor but omit it on the history path would silently lose
-    // pending non-CLIENT approval cards on every reload/reconnect refetch
-    // (they also ignore `escalatedApprovals`). Realtime/history parity is
-    // opt-in: pass the same explicit list to both.
-    displayApprovalTypes,
-    batchApprovalsEnabled,
-  } = options
-
-  const processedMessages: ProcessedMessage[] = []
-  const accumulator = createMessageSegmentAccumulator({ onApprove, onReject })
-  const escalatedApprovals = new Map<
-    string,
-    { command: string; explanation?: string; approvalType: string; toolCalls?: PendingToolCallData[] }
-  >()
-  
-  let currentAssistantId: string | null = null
-  let currentAssistantTimestamp: Date | null = null
-  let lastAssistantId: string | null = null
-  // MAX persisted seq across the rows grouped into the current assistant turn
-  // — carried onto the flushed message's streamSeq for per-role merge coverage.
-  let currentAssistantStreamSeq: number | undefined
-
-  /**
-   * Flush the current assistant message to processedMessages.
-   * Uses the LAST message ID in the group for stable React keys across page boundaries.
-   */
-  const flushAssistantMessage = () => {
-    const idToUse = lastAssistantId || currentAssistantId
-    if (idToUse && accumulator.hasContent()) {
-      processedMessages.push({
-        id: idToUse,
-        role: 'assistant',
-        content: accumulator.getSegments(),
-        name: assistantName,
-        assistantType,
-        authorType: assistantType as AuthorType,
-        timestamp: currentAssistantTimestamp || new Date(),
-        avatar: assistantAvatar,
-        ...(currentAssistantStreamSeq !== undefined ? { streamSeq: currentAssistantStreamSeq } : {}),
-      })
-      accumulator.resetSegments()
-    }
-    // Reset grouping identity + seq UNCONDITIONALLY — even on an EMPTY flush (an
-    // assistant turn whose only data was a filtered/escalated approval renders
-    // nothing, so `hasContent()` is false and nothing is pushed). Left inside
-    // the push-block, a stale id/timestamp and (worse) a stale
-    // `currentAssistantStreamSeq` bleed into the NEXT assistant turn: `!currentAssistantId`
-    // stays false so it keeps the old id/timestamp, and `Math.max` inflates its
-    // streamSeq — which then over-covers synthetics in the history merge.
-    currentAssistantId = null
-    currentAssistantTimestamp = null
-    lastAssistantId = null
-    currentAssistantStreamSeq = undefined
-  }
-
-  messages.forEach((msg, index) => {
-    // Filter by chat type if specified
-    if (chatTypeFilter && msg.chatType !== chatTypeFilter) return
-
-    const messageDataArray = Array.isArray(msg.messageData)
-      ? msg.messageData
-      : msg.messageData
-      ? [msg.messageData]
-      : []
-
-    const hasStandaloneData = messageDataArray.some((data) =>
-      data.type === MESSAGE_TYPE.SYSTEM
-    )
-    if (hasStandaloneData) {
-      flushAssistantMessage()
-      pushStandaloneMessages(processedMessages, msg, messageDataArray)
-      return
-    }
-
-    const isUserMessage =
-      msg.owner?.type === OWNER_TYPE.CLIENT || msg.owner?.type === OWNER_TYPE.ADMIN
-
-    if (isUserMessage) {
-      flushAssistantMessage()
-
-      const userAuthorType: AuthorType = msg.owner?.type === OWNER_TYPE.ADMIN ? 'admin' : 'user'
-      messageDataArray.forEach((data) => {
-        if (data.type === MESSAGE_TYPE.TEXT && 'text' in data && data.text) {
-          // `TextData.contextItems` (server: `[{ type, id }]`) — the entity
-          // context the user attached to this message. Surface it so the bubble
-          // renders its chip strip from history (no label on the wire → fall
-          // back to the id, matching the realtime path).
-          const rawContext = (data as { contextItems?: Array<{ type?: unknown; id?: unknown }> }).contextItems
-          const contextItems = Array.isArray(rawContext)
-            ? rawContext
-                .filter((c) => typeof c?.type === 'string' && typeof c?.id === 'string')
-                .map((c) => ({ type: c.type as string, id: c.id as string, label: c.id as string }))
-            : undefined
-          processedMessages.push({
-            id: msg.id,
-            role: 'user',
-            content: data.text,
-            name: getOwnerDisplayName(msg.owner),
-            avatar: getOwnerAvatar(msg.owner),
-            authorType: userAuthorType,
-            timestamp: new Date(msg.createdAt),
-            ...(contextItems && contextItems.length > 0 ? { contextItems } : {}),
-            ...(typeof msg.lastChunkStreamSeq === 'number' ? { streamSeq: msg.lastChunkStreamSeq } : {}),
-          })
-        }
-      })
-    } else {
-      if (!currentAssistantId) {
-        currentAssistantId = msg.id
-        currentAssistantTimestamp = new Date(msg.createdAt)
-      }
-      lastAssistantId = msg.id
-      if (typeof msg.lastChunkStreamSeq === 'number') {
-        currentAssistantStreamSeq =
-          currentAssistantStreamSeq === undefined
-            ? msg.lastChunkStreamSeq
-            : Math.max(currentAssistantStreamSeq, msg.lastChunkStreamSeq)
-      }
-
-      messageDataArray.forEach((data) => {
-        processMessageData(
-          data,
-          accumulator,
-          approvalStatuses,
-          { displayApprovalTypes, batchApprovalsEnabled },
-          escalatedApprovals,
-        )
-      })
-
-      // Check if we should flush (next message is from user or last message)
-      const nextMsg = messages[index + 1]
-      const isLastMessage = index === messages.length - 1
-      const nextIsFromUser =
-        nextMsg &&
-        (nextMsg.owner?.type === OWNER_TYPE.CLIENT || nextMsg.owner?.type === OWNER_TYPE.ADMIN)
-
-      if (isLastMessage || nextIsFromUser) {
-        flushAssistantMessage()
-      }
-    }
-  })
-
-  flushAssistantMessage()
-
-  const pendingApprovalSegments = accumulator.flushPendingApprovals()
-  if (pendingApprovalSegments.length > 0) {
-    processedMessages.push({
-      id: `pending-approvals-${Date.now()}`,
-      role: 'assistant',
-      content: pendingApprovalSegments,
-      name: assistantName,
-      assistantType,
-      timestamp: new Date(),
-      avatar: assistantAvatar,
-    })
-  }
-
-  return {
-    messages: processedMessages,
-    escalatedApprovals: escalatedApprovals
-  }
-}
-
-/**
- * Process a single message data item into segments
- */
-function processMessageData(
-  data: MessageData,
-  accumulator: MessageSegmentAccumulator,
-  approvalStatuses: Record<string, string>,
-  options: MessageProcessingOptions = {},
-  escalatedApprovals?: Map<
-    string,
-    { command: string; explanation?: string; approvalType: string; toolCalls?: PendingToolCallData[] }
-  >
-): void {
-  // batchApprovalsEnabled is owned by the consumer (oss-tenant chat client /
-  // openframe-frontend tickets). Defaults to ON so consumers that haven't
-  // wired the flag yet get the batch UI; pass `false` to force legacy.
-  const { displayApprovalTypes, batchApprovalsEnabled = true } = options
+export function decodeHistoricalMessageData(data: MessageData): ChatStreamEvent | null {
   switch (data.type) {
     case MESSAGE_TYPE.TEXT:
       if ('text' in data && data.text) {
-        accumulator.appendText(data.text)
+        return { type: 'text-delta', text: data.text }
       }
-      break
+      return null
 
     case MESSAGE_TYPE.THINKING:
       if ('text' in data && data.text) {
-        accumulator.appendThinking(data.text)
+        return { type: 'thinking-delta', text: data.text }
       }
-      break
+      return null
 
     case MESSAGE_TYPE.EXECUTING_TOOL:
       if ('integratedToolType' in data) {
-        accumulator.addToolExecution({
-          type: 'tool_execution',
+        return {
+          type: 'tool-execution',
           data: {
             type: 'EXECUTING_TOOL',
             integratedToolType: data.integratedToolType || '',
@@ -287,14 +123,14 @@ function processMessageData(
             parameters: data.parameters,
             toolExecutionRequestId: data.toolExecutionRequestId,
           },
-        })
+        }
       }
-      break
+      return null
 
     case MESSAGE_TYPE.EXECUTED_TOOL:
       if ('integratedToolType' in data) {
-        accumulator.addToolExecution({
-          type: 'tool_execution',
+        return {
+          type: 'tool-execution',
           data: {
             type: 'EXECUTED_TOOL',
             integratedToolType: data.integratedToolType || '',
@@ -305,218 +141,294 @@ function processMessageData(
             success: data.success,
             toolExecutionRequestId: data.toolExecutionRequestId,
           },
-        })
+        }
       }
-      break
+      return null
 
     case MESSAGE_TYPE.APPROVAL_REQUEST:
       if ('approvalRequestId' in data && data.approvalRequestId) {
-        const approvalType = data.approvalType || 'CLIENT'
-        const toolCalls: PendingToolCallData[] | undefined = Array.isArray(data.toolCalls)
-          ? data.toolCalls
-          : undefined
-        const isBatch = !!toolCalls && toolCalls.length > 0
-
-        if (!displayApprovalTypes || displayApprovalTypes.includes(approvalType)) {
-          if (isBatch) {
-            const status = (approvalStatuses[data.approvalRequestId] as ChatApprovalStatus) || 'pending'
-            if (batchApprovalsEnabled) {
-              accumulator.addApprovalBatch(data.approvalRequestId, approvalType, toolCalls!, status)
-            } else {
-              // Flag OFF — unfold batch into N legacy approval cards (same id).
-              for (const call of toolCalls!) {
-                if (!call.requiresApproval) continue
-                accumulator.addApprovalRequest(
-                  data.approvalRequestId,
-                  getCommandText(call),
-                  call.toolExplanation,
-                  approvalType,
-                  status,
-                )
-              }
-            }
-          } else {
-            // The resolution may already be known to the consumer (realtime
-            // flipped it and fed it back via `approvalStatuses`) while the
-            // matching APPROVAL_RESULT row is absent from the fetched history
-            // pages. Without this, the single-approval path tracks it as
-            // pending and `flushPendingApprovals()` resurrects it as a stale
-            // sticky card on every history re-process. Mirror the batch path
-            // and honor `approvalStatuses`.
-            const resolvedStatus = approvalStatuses[data.approvalRequestId] as ChatApprovalStatus | undefined
-            if (resolvedStatus === 'approved' || resolvedStatus === 'rejected') {
-              accumulator.addApprovalRequest(
-                data.approvalRequestId,
-                data.command || '',
-                data.explanation,
-                approvalType,
-                resolvedStatus,
-              )
-            } else {
-              accumulator.trackApprovalRequest(data.approvalRequestId, {
-                command: data.command || '',
-                explanation: data.explanation,
-                approvalType,
-              })
-            }
-          }
-        } else {
-          escalatedApprovals?.set(data.approvalRequestId, {
-            command: data.command || '',
-            explanation: data.explanation,
-            approvalType,
-            ...(isBatch ? { toolCalls } : {}),
-          })
+        return {
+          type: 'approval-request',
+          requestId: data.approvalRequestId,
+          approvalType: data.approvalType || 'CLIENT',
+          command: data.command || '',
+          explanation: data.explanation,
+          ...(Array.isArray(data.toolCalls) ? { toolCalls: data.toolCalls } : {}),
         }
       }
-      break
+      return null
 
     case MESSAGE_TYPE.APPROVAL_RESULT:
       if ('approvalRequestId' in data && data.approvalRequestId) {
-        const existingStatus = approvalStatuses[data.approvalRequestId] as ChatApprovalStatus | undefined
-        const status: ChatApprovalStatus = existingStatus || (data.approved ? 'approved' : 'rejected')
-        const resolvedByName = 'resolvedByName' in data ? data.resolvedByName : undefined
-        const escalatedData = escalatedApprovals?.get(data.approvalRequestId)
+        return {
+          type: 'approval-resolved',
+          requestId: data.approvalRequestId,
+          status: data.approved ? 'approved' : 'rejected',
+          approvalType: data.approvalType,
+          resolvedByName: 'resolvedByName' in data ? data.resolvedByName : undefined,
+        }
+      }
+      return null
+
+    case MESSAGE_TYPE.ERROR:
+      if ('error' in data) {
+        return {
+          type: 'error',
+          title: data.error || 'An error occurred',
+          details: 'details' in data ? data.details : undefined,
+        }
+      }
+      return null
+
+    case MESSAGE_TYPE.CONTEXT_COMPACTION_START:
+      return { type: 'compaction', phase: 'start' }
+
+    case MESSAGE_TYPE.CONTEXT_COMPACTION_END:
+      return {
+        type: 'compaction',
+        phase: 'end',
+        summary: 'summary' in data && typeof data.summary === 'string' ? data.summary : undefined,
+      }
+
+    case MESSAGE_TYPE.SYSTEM:
+      if ('text' in data && data.text) {
+        return { type: 'participant', kind: 'system', text: data.text }
+      }
+      return null
+
+    default:
+      // Unknown message type — ignore.
+      return null
+  }
+}
+
+// =============================================================================
+// Replay — decoded event → per-turn kernel (history approval semantics)
+// =============================================================================
+
+type EscalatedApprovals = Map<
+  string,
+  { command: string; explanation?: string; approvalType: string; toolCalls?: PendingToolCallData[] }
+>
+
+/**
+ * Replay one decoded event into the shared per-turn segment kernel with the
+ * HISTORY approval semantics (an omitted `displayApprovalTypes` means
+ * "display every approval type" — the original history behavior; the
+ * realtime reducer defaults to `['CLIENT']`).
+ */
+function applyHistoryEvent(
+  event: ChatStreamEvent,
+  accumulator: MessageSegmentAccumulator,
+  approvalStatuses: Record<string, string>,
+  options: MessageProcessingOptions,
+  escalatedApprovals?: EscalatedApprovals,
+): void {
+  // batchApprovalsEnabled is owned by the consumer (oss-tenant chat client /
+  // openframe-frontend tickets). Defaults to ON so consumers that haven't
+  // wired the flag yet get the batch UI; pass `false` to force legacy.
+  const { displayApprovalTypes, batchApprovalsEnabled = true } = options
+
+  switch (event.type) {
+    case 'text-delta':
+      accumulator.appendText(event.text)
+      break
+
+    case 'thinking-delta':
+      accumulator.appendThinking(event.text)
+      break
 
-        if (escalatedData?.toolCalls && escalatedData.toolCalls.length > 0) {
+    case 'tool-execution':
+      accumulator.addToolExecution({ type: 'tool_execution', data: event.data })
+      break
+
+    case 'approval-request': {
+      const approvalType = event.approvalType || 'CLIENT'
+      const toolCalls = event.toolCalls as PendingToolCallData[] | undefined
+      const isBatch = !!toolCalls && toolCalls.length > 0
+
+      if (!displayApprovalTypes || displayApprovalTypes.includes(approvalType)) {
+        if (isBatch) {
+          const status = (approvalStatuses[event.requestId] as ChatApprovalStatus) || 'pending'
           if (batchApprovalsEnabled) {
-            accumulator.addApprovalBatch(
-              data.approvalRequestId,
-              escalatedData.approvalType,
-              escalatedData.toolCalls,
-              status,
-              undefined,
-              resolvedByName,
-            )
+            accumulator.addApprovalBatch(event.requestId, approvalType, toolCalls!, status)
           } else {
-            for (const call of escalatedData.toolCalls) {
+            // Flag OFF — unfold batch into N legacy approval cards (same id).
+            for (const call of toolCalls!) {
               if (!call.requiresApproval) continue
               accumulator.addApprovalRequest(
-                data.approvalRequestId,
+                event.requestId,
                 getCommandText(call),
                 call.toolExplanation,
-                escalatedData.approvalType,
+                approvalType,
                 status,
               )
             }
           }
-          escalatedApprovals?.delete(data.approvalRequestId)
-          break
-        }
-
-        if (escalatedData) {
-          accumulator.trackApprovalRequest(data.approvalRequestId, {
-            command: escalatedData.command,
-            explanation: escalatedData.explanation,
-            approvalType: escalatedData.approvalType,
-          })
-          escalatedApprovals?.delete(data.approvalRequestId)
+        } else {
+          // The resolution may already be known to the consumer (realtime
+          // flipped it and fed it back via `approvalStatuses`) while the
+          // matching APPROVAL_RESULT row is absent from the fetched history
+          // pages. Without this, the single-approval path tracks it as
+          // pending and `flushPendingApprovals()` resurrects it as a stale
+          // sticky card on every history re-process. Mirror the batch path
+          // and honor `approvalStatuses`.
+          const resolvedStatus = approvalStatuses[event.requestId] as ChatApprovalStatus | undefined
+          if (resolvedStatus === 'approved' || resolvedStatus === 'rejected') {
+            accumulator.addApprovalRequest(
+              event.requestId,
+              event.command || '',
+              event.explanation,
+              approvalType,
+              resolvedStatus,
+            )
+          } else {
+            accumulator.trackApprovalRequest(event.requestId, {
+              command: event.command || '',
+              explanation: event.explanation,
+              approvalType,
+            })
+          }
         }
-
-        // If a segment with this id is already present (batch or legacy), just flip its status.
-        // updateApprovalStatus matches both `approval_batch` and `approval_request` segments.
-        const before = accumulator.getSegments()
-        const after = accumulator.updateApprovalStatus(data.approvalRequestId, status, resolvedByName)
-        const updatedExisting = before.some((s, i) => after[i] !== s)
-        if (updatedExisting) break
-
-        accumulator.processApprovalResult(
-          data.approvalRequestId,
-          status === 'approved',
-          data.approvalType || 'USER'
-        )
+      } else {
+        escalatedApprovals?.set(event.requestId, {
+          command: event.command || '',
+          explanation: event.explanation,
+          approvalType,
+          ...(isBatch ? { toolCalls } : {}),
+        })
       }
       break
+    }
 
-    case MESSAGE_TYPE.ERROR:
-      if ('error' in data) {
-        let message: string | undefined
-        if ('details' in data && data?.details) {
-          try {
-            message = JSON.parse(data.details)?.error?.message
-          } catch {
-            message = data.details
+    case 'approval-resolved': {
+      const requestId = event.requestId
+      if (!requestId) break
+      const existingStatus = approvalStatuses[requestId] as ChatApprovalStatus | undefined
+      const status: ChatApprovalStatus = existingStatus || event.status
+      const resolvedByName = event.resolvedByName
+      const escalatedData = escalatedApprovals?.get(requestId)
+
+      if (escalatedData?.toolCalls && escalatedData.toolCalls.length > 0) {
+        if (batchApprovalsEnabled) {
+          accumulator.addApprovalBatch(
+            requestId,
+            escalatedData.approvalType,
+            escalatedData.toolCalls,
+            status,
+            undefined,
+            resolvedByName,
+          )
+        } else {
+          for (const call of escalatedData.toolCalls) {
+            if (!call.requiresApproval) continue
+            accumulator.addApprovalRequest(
+              requestId,
+              getCommandText(call),
+              call.toolExplanation,
+              escalatedData.approvalType,
+              status,
+            )
           }
         }
-        accumulator.addError(
-          data.error || 'An error occurred',
-          message
-        )
+        escalatedApprovals?.delete(requestId)
+        break
       }
-      break
 
-    case MESSAGE_TYPE.CONTEXT_COMPACTION_START:
-      accumulator.addContextCompaction()
+      if (escalatedData) {
+        accumulator.trackApprovalRequest(requestId, {
+          command: escalatedData.command,
+          explanation: escalatedData.explanation,
+          approvalType: escalatedData.approvalType,
+        })
+        escalatedApprovals?.delete(requestId)
+      }
+
+      // If a segment with this id is already present (batch or legacy), just
+      // flip its status. updateApprovalStatus matches both `approval_batch`
+      // and `approval_request` segments.
+      const before = accumulator.getSegments()
+      const after = accumulator.updateApprovalStatus(requestId, status, resolvedByName)
+      const updatedExisting = before.some((s, i) => after[i] !== s)
+      if (updatedExisting) break
+
+      accumulator.processApprovalResult(
+        requestId,
+        status === 'approved',
+        event.approvalType || 'USER',
+      )
       break
+    }
 
-    case MESSAGE_TYPE.CONTEXT_COMPACTION_END: {
-      const summary = 'summary' in data && typeof data.summary === 'string' ? data.summary : undefined
-      accumulator.completeContextCompaction(summary)
+    case 'error': {
+      let message: string | undefined
+      if (event.details) {
+        try {
+          message = JSON.parse(event.details)?.error?.message
+        } catch {
+          message = event.details
+        }
+      }
+      accumulator.addError(event.title, message)
       break
     }
 
+    case 'compaction':
+      if (event.phase === 'start') {
+        accumulator.addContextCompaction()
+      } else {
+        accumulator.completeContextCompaction(event.summary)
+      }
+      break
+
     default:
-      // Unknown message type - ignore
+      // Participant/system rows are envelope concerns; everything else is
+      // realtime-only vocabulary that never appears in persisted rows.
       break
   }
 }
 
+// =============================================================================
+// Envelope — ONE implementation (previous duplicate deleted)
+// =============================================================================
+
 /**
- * Extract error messages from historical messages
- * Returns a separate array of error messages that should be displayed
+ * Result type for historical message processing
  */
-export function extractErrorMessages(
-  messages: HistoricalMessage[],
-  options: MessageProcessingOptions = {}
-): ProcessedMessage[] {
-  const { assistantName = 'Fae', assistantType = 'fae', assistantAvatar, chatTypeFilter } = options
-
-  const errorMessages: ProcessedMessage[] = []
-
-  messages.forEach((msg) => {
-    if (chatTypeFilter && msg.chatType !== chatTypeFilter) return
-
-    const messageDataArray = Array.isArray(msg.messageData)
-      ? msg.messageData
-      : msg.messageData
-      ? [msg.messageData]
-      : []
-
-    messageDataArray.forEach((data) => {
-      if (data.type === MESSAGE_TYPE.ERROR) {
-        errorMessages.push({
-          id: `${msg.id}-error`,
-          role: 'error',
-          content: 'error' in data && data.error ? data.error : 'An error occurred',
-          name: assistantName,
-          assistantType,
-          timestamp: new Date(msg.createdAt),
-          avatar: assistantAvatar,
-        })
-      }
-    })
-  })
-
-  return errorMessages
+export interface ProcessHistoricalMessagesResult {
+  messages: ProcessedMessage[]
+  escalatedApprovals: EscalatedApprovals
 }
 
 /**
- * Process messages and include error messages in the correct order
+ * Process an array of historical messages into display-ready format
  */
-export function processHistoricalMessagesWithErrors(
+export function processHistoricalMessages(
   messages: HistoricalMessage[],
-  options: MessageProcessingOptions = {}
+  options: MessageProcessingOptions = {},
 ): ProcessHistoricalMessagesResult {
-  // displayApprovalTypes omitted = display every approval type (original
-  // history semantics — see the matching note in processHistoricalMessages).
-  const { chatTypeFilter, assistantName = 'Fae', assistantType = 'fae', assistantAvatar, onApprove, onReject, approvalStatuses = {}, displayApprovalTypes, batchApprovalsEnabled } = options
+  const {
+    assistantName = 'Fae',
+    assistantType = 'fae',
+    assistantAvatar,
+    onApprove,
+    onReject,
+    chatTypeFilter,
+    approvalStatuses = {},
+    // An omitted option means "display every approval type" — the original
+    // history semantics. Deliberately NOT defaulted to the realtime
+    // reducer's ['CLIENT']: consumers that pass a wider list to their
+    // realtime path but omit it on the history path would silently lose
+    // pending non-CLIENT approval cards on every reload/reconnect refetch
+    // (they also ignore `escalatedApprovals`). Realtime/history parity is
+    // opt-in: pass the same explicit list to both.
+    displayApprovalTypes,
+    batchApprovalsEnabled,
+  } = options
 
   const processedMessages: ProcessedMessage[] = []
   const accumulator = createMessageSegmentAccumulator({ onApprove, onReject })
-  const escalatedApprovals = new Map<
-    string,
-    { command: string; explanation?: string; approvalType: string; toolCalls?: PendingToolCallData[] }
-  >()
+  const escalatedApprovals: EscalatedApprovals = new Map()
 
   let currentAssistantId: string | null = null
   let currentAssistantTimestamp: Date | null = null
@@ -525,6 +437,10 @@ export function processHistoricalMessagesWithErrors(
   // — carried onto the flushed message's streamSeq for per-role merge coverage.
   let currentAssistantStreamSeq: number | undefined
 
+  /**
+   * Flush the current assistant message to processedMessages.
+   * Uses the LAST message ID in the group for stable React keys across page boundaries.
+   */
   const flushAssistantMessage = () => {
     const idToUse = lastAssistantId || currentAssistantId
     if (idToUse && accumulator.hasContent()) {
@@ -555,6 +471,7 @@ export function processHistoricalMessagesWithErrors(
   }
 
   messages.forEach((msg, index) => {
+    // Filter by chat type if specified
     if (chatTypeFilter && msg.chatType !== chatTypeFilter) return
 
     const messageDataArray = Array.isArray(msg.messageData)
@@ -618,8 +535,10 @@ export function processHistoricalMessagesWithErrors(
       }
 
       messageDataArray.forEach((data) => {
-        processMessageData(
-          data,
+        const event = decodeHistoricalMessageData(data)
+        if (!event) return
+        applyHistoryEvent(
+          event,
           accumulator,
           approvalStatuses,
           { displayApprovalTypes, batchApprovalsEnabled },
@@ -627,6 +546,7 @@ export function processHistoricalMessagesWithErrors(
         )
       })
 
+      // Check if we should flush (next message is from user or last message)
       const nextMsg = messages[index + 1]
       const isLastMessage = index === messages.length - 1
       const nextIsFromUser =
@@ -659,3 +579,52 @@ export function processHistoricalMessagesWithErrors(
     escalatedApprovals: escalatedApprovals
   }
 }
+
+/**
+ * Extract error messages from historical messages
+ * Returns a separate array of error messages that should be displayed
+ */
+export function extractErrorMessages(
+  messages: HistoricalMessage[],
+  options: MessageProcessingOptions = {}
+): ProcessedMessage[] {
+  const { assistantName = 'Fae', assistantType = 'fae', assistantAvatar, chatTypeFilter } = options
+
+  const errorMessages: ProcessedMessage[] = []
+
+  messages.forEach((msg) => {
+    if (chatTypeFilter && msg.chatType !== chatTypeFilter) return
+
+    const messageDataArray = Array.isArray(msg.messageData)
+      ? msg.messageData
+      : msg.messageData
+      ? [msg.messageData]
+      : []
+
+    messageDataArray.forEach((data) => {
+      if (data.type === MESSAGE_TYPE.ERROR) {
+        errorMessages.push({
+          id: `${msg.id}-error`,
+          role: 'error',
+          content: 'error' in data && data.error ? data.error : 'An error occurred',
+          name: assistantName,
+          assistantType,
+          timestamp: new Date(msg.createdAt),
+          avatar: assistantAvatar,
+        })
+      }
+    })
+  })
+
+  return errorMessages
+}
+
+/**
+ * Process messages and include error messages in the correct order.
+ *
+ * HISTORICAL ALIAS: this was a byte-identical second copy of
+ * `processHistoricalMessages` (both goldens snapshot identical output on the
+ * same corpus). The duplicate envelope was deleted in Phase 3 — kept as an
+ * alias for the established import sites.
+ */
+export const processHistoricalMessagesWithErrors = processHistoricalMessages