fix(agents): preserve openai replay ids and timeout hooks

src/agents/pi-embedded-runner.sanitize-session-history.test-harness.ts

@@ -123,7 +123,7 @@ export function expectOpenAIResponsesStrictSanitizeCall(
     "session:history",
     expect.objectContaining({
       sanitizeMode: "images-only",
-      sanitizeToolCallIds: true,
+      sanitizeToolCallIds: false,
       toolCallIdMode: "strict",
     }),
   );

src/agents/pi-embedded-runner/replay-history.ts

@@ -28,12 +28,12 @@ import {
   sanitizeToolUseResultPairing,
   stripToolResultDetails,
 } from "../session-transcript-repair.js";
+import { sanitizeToolCallIdsForCloudCodeAssist } from "../tool-call-id.js";
 import type { TranscriptPolicy } from "../transcript-policy.js";
 import {
   resolveTranscriptPolicy,
   shouldAllowProviderOwnedThinkingReplay,
 } from "../transcript-policy.js";
-import { sanitizeToolCallIdsForCloudCodeAssist } from "../tool-call-id.js";
 import {
   makeZeroUsageSnapshot,
   normalizeUsage,
@@ -408,12 +408,17 @@ export async function sanitizeSessionHistory(params: {
     modelApi: params.modelApi,
     policy,
   });
+  const isOpenAIResponsesApi =
+    params.modelApi === "openai-responses" ||
+    params.modelApi === "openai-codex-responses" ||
+    params.modelApi === "azure-openai-responses";
   const sanitizedImages = await sanitizeSessionMessagesImages(
     withInterSessionMarkers,
     "session:history",
     {
       sanitizeMode: policy.sanitizeMode,
-      sanitizeToolCallIds: policy.sanitizeToolCallIds && !allowProviderOwnedThinkingReplay,
+      sanitizeToolCallIds:
+        policy.sanitizeToolCallIds && !allowProviderOwnedThinkingReplay && !isOpenAIResponsesApi,
       toolCallIdMode: policy.toolCallIdMode,
       preserveNativeAnthropicToolUseIds: policy.preserveNativeAnthropicToolUseIds,
       preserveSignatures: policy.preserveSignatures,
@@ -429,7 +434,7 @@ export async function sanitizeSessionHistory(params: {
     allowProviderOwnedThinkingReplay,
   });
   const sanitizedToolIds =
-    policy.sanitizeToolCallIds && policy.toolCallIdMode
+    policy.sanitizeToolCallIds && policy.toolCallIdMode && !isOpenAIResponsesApi
       ? sanitizeToolCallIdsForCloudCodeAssist(sanitizedToolCalls, policy.toolCallIdMode, {
           preserveNativeAnthropicToolUseIds: policy.preserveNativeAnthropicToolUseIds,
           preserveReplaySafeThinkingToolCallIds: allowProviderOwnedThinkingReplay,
@@ -446,10 +451,6 @@ export async function sanitizeSessionHistory(params: {
     stripStaleAssistantUsageBeforeLatestCompaction(sanitizedToolResults),
   );
 
-  const isOpenAIResponsesApi =
-    params.modelApi === "openai-responses" ||
-    params.modelApi === "openai-codex-responses" ||
-    params.modelApi === "azure-openai-responses";
   const hasSnapshot = Boolean(params.provider || params.modelApi || params.modelId);
   const priorSnapshot = hasSnapshot ? readLastModelSnapshot(params.sessionManager) : null;
   const modelChanged = priorSnapshot

src/agents/pi-embedded-runner/run.overflow-compaction.harness.ts

@@ -484,6 +484,10 @@ export async function loadRunOverflowCompactionHarness(): Promise<{
     buildEmbeddedRunPayloads: vi.fn(() => []),
   }));
 
+  vi.doMock("./compaction-hooks.js", () => ({
+    runPostCompactionSideEffects: mockedRunPostCompactionSideEffects,
+  }));
+
   vi.doMock("./compact.js", () => ({
     runPostCompactionSideEffects: mockedRunPostCompactionSideEffects,
   }));

src/agents/pi-embedded-runner/run/attempt.ts

@@ -1150,7 +1150,16 @@ export async function runEmbeddedAttempt(
       // historical messages at attempt start, but the agent loop's internal tool call →
       // tool result cycles bypass that path. Wrap streamFn so every outbound request
       // sees sanitized tool call IDs.
-      if (transcriptPolicy.sanitizeToolCallIds && transcriptPolicy.toolCallIdMode) {
+      const isOpenAIResponsesApi =
+        params.model.api === "openai-responses" ||
+        params.model.api === "azure-openai-responses" ||
+        params.model.api === "openai-codex-responses";
+
+      if (
+        transcriptPolicy.sanitizeToolCallIds &&
+        transcriptPolicy.toolCallIdMode &&
+        !isOpenAIResponsesApi
+      ) {
         const inner = activeSession.agent.streamFn;
         const mode = transcriptPolicy.toolCallIdMode;
         activeSession.agent.streamFn = (model, context, options) => {
@@ -1183,11 +1192,7 @@ export async function runEmbeddedAttempt(
         };
       }
 
-      if (
-        params.model.api === "openai-responses" ||
-        params.model.api === "azure-openai-responses" ||
-        params.model.api === "openai-codex-responses"
-      ) {
+      if (isOpenAIResponsesApi) {
         const inner = activeSession.agent.streamFn;
         activeSession.agent.streamFn = (model, context, options) => {
           const ctx = context as unknown as { messages?: unknown };