fix(agent): harden empty attempt retry handling

src/agents/pi-embedded-runner/run.empty-error-retry.test.ts

@@ -112,14 +112,15 @@ describe("runEmbeddedPiAgent silent-error retry", () => {
 
   it("does not retry when stopReason=stop and output=0 (out of scope)", async () => {
     // Clean stop with no output is a legitimate silent reply (e.g. NO_REPLY
-    // token path), not a crash. This retry must not trigger there.
+    // token path), not a crash. Use a plain provider/model so this test stays
+    // scoped to the silent-error retry instead of the empty-response retry.
     mockedRunEmbeddedAttempt.mockResolvedValueOnce(
       makeAttemptResult({
         assistantTexts: [],
         lastAssistant: {
           stopReason: "stop",
-          provider: "ollama",
-          model: "glm-5.1:cloud",
+          provider: "plain-provider",
+          model: "plain-model",
           content: [],
           usage: { input: 100, output: 0, totalTokens: 100 },
         } as unknown as EmbeddedRunAttemptResult["lastAssistant"],
@@ -128,8 +129,8 @@ describe("runEmbeddedPiAgent silent-error retry", () => {
 
     await runEmbeddedPiAgent({
       ...overflowBaseRunParams,
-      provider: "ollama",
-      model: "glm-5.1:cloud",
+      provider: "plain-provider",
+      model: "plain-model",
       runId: "run-empty-error-retry-skip-clean-stop",
     });
 

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

@@ -113,6 +113,7 @@ import {
   DEFAULT_EMPTY_RESPONSE_RETRY_LIMIT,
   DEFAULT_REASONING_ONLY_RETRY_LIMIT,
   resolveAckExecutionFastPathInstruction,
+  resolveAttemptReplayMetadata,
   extractPlanningOnlyPlanDetails,
   resolveEmptyResponseRetryInstruction,
   resolveIncompleteTurnPayloadText,
@@ -151,6 +152,36 @@ import { createUsageAccumulator, mergeUsageIntoAccumulator } from "./usage-accum
 type ApiKeyInfo = ResolvedProviderAuth;
 
 const MAX_SAME_MODEL_IDLE_TIMEOUT_RETRIES = 1;
+type EmbeddedRunAttemptForRunner = Awaited<ReturnType<typeof runEmbeddedAttemptWithBackend>>;
+
+function normalizeEmbeddedRunAttemptResult(
+  attempt: EmbeddedRunAttemptForRunner,
+): EmbeddedRunAttemptForRunner {
+  const raw = attempt as EmbeddedRunAttemptForRunner & {
+    assistantTexts?: EmbeddedRunAttemptForRunner["assistantTexts"] | null;
+    toolMetas?: EmbeddedRunAttemptForRunner["toolMetas"] | null;
+    messagesSnapshot?: EmbeddedRunAttemptForRunner["messagesSnapshot"] | null;
+    messagingToolSentTexts?: EmbeddedRunAttemptForRunner["messagingToolSentTexts"] | null;
+    messagingToolSentMediaUrls?: EmbeddedRunAttemptForRunner["messagingToolSentMediaUrls"] | null;
+    messagingToolSentTargets?: EmbeddedRunAttemptForRunner["messagingToolSentTargets"] | null;
+    itemLifecycle?: EmbeddedRunAttemptForRunner["itemLifecycle"] | null;
+  };
+  return {
+    ...attempt,
+    assistantTexts: raw.assistantTexts ?? [],
+    toolMetas: raw.toolMetas ?? [],
+    messagesSnapshot: raw.messagesSnapshot ?? [],
+    messagingToolSentTexts: raw.messagingToolSentTexts ?? [],
+    messagingToolSentMediaUrls: raw.messagingToolSentMediaUrls ?? [],
+    messagingToolSentTargets: raw.messagingToolSentTargets ?? [],
+    itemLifecycle: raw.itemLifecycle ?? {
+      startedCount: 0,
+      completedCount: 0,
+      activeCount: 0,
+    },
+    replayMetadata: resolveAttemptReplayMetadata(raw),
+  };
+}
 
 function createEmptyAuthProfileStore(): AuthProfileStore {
   return {
@@ -855,7 +886,7 @@ export async function runEmbeddedPiAgent(
             },
           });
 
-          const attempt = await runEmbeddedAttemptWithBackend({
+          const rawAttempt = await runEmbeddedAttemptWithBackend({
             sessionId: activeSessionId,
             sessionKey: resolvedSessionKey,
             sandboxSessionKey: params.sandboxSessionKey,
@@ -960,6 +991,7 @@ export async function runEmbeddedPiAgent(
             bootstrapPromptWarningSignature:
               bootstrapPromptWarningSignaturesSeen[bootstrapPromptWarningSignaturesSeen.length - 1],
           });
+          const attempt = normalizeEmbeddedRunAttemptResult(rawAttempt);
 
           const {
             aborted,

src/agents/pi-embedded-runner/run/incomplete-turn.ts

@@ -72,6 +72,11 @@ type RunLivenessAttempt = Pick<
   "lastAssistant" | "promptErrorSource" | "replayMetadata" | "timedOutDuringCompaction"
 >;
 
+const REPLAY_UNSAFE_FALLBACK_METADATA: EmbeddedRunAttemptResult["replayMetadata"] = {
+  hadPotentialSideEffects: true,
+  replaySafe: false,
+};
+
 export function isIncompleteTerminalAssistantTurn(params: {
   hasAssistantVisibleText: boolean;
   lastAssistant?: { stopReason?: string } | null;
@@ -211,6 +216,12 @@ export function buildAttemptReplayMetadata(
   };
 }
 
+export function resolveAttemptReplayMetadata(attempt: {
+  replayMetadata?: EmbeddedRunAttemptResult["replayMetadata"] | null;
+}): EmbeddedRunAttemptResult["replayMetadata"] {
+  return attempt.replayMetadata ?? REPLAY_UNSAFE_FALLBACK_METADATA;
+}
+
 export function resolveIncompleteTurnPayloadText(params: {
   payloadCount: number;
   aborted: boolean;
@@ -258,7 +269,7 @@ export function resolveIncompleteTurnPayloadText(params: {
     return null;
   }
 
-  return params.attempt.replayMetadata.hadPotentialSideEffects
+  return resolveAttemptReplayMetadata(params.attempt).hadPotentialSideEffects
     ? "⚠️ Agent couldn't generate a response. Note: some tool actions may have already been executed — please verify before retrying."
     : "⚠️ Agent couldn't generate a response. Please try again.";
 }
@@ -351,7 +362,7 @@ export function resolveReplayInvalidFlag(params: {
   incompleteTurnText?: string | null;
 }): boolean {
   return (
-    !params.attempt.replayMetadata.replaySafe ||
+    !resolveAttemptReplayMetadata(params.attempt).replaySafe ||
     params.attempt.promptErrorSource === "compaction" ||
     params.attempt.timedOutDuringCompaction ||
     Boolean(params.incompleteTurnText)
@@ -465,7 +476,7 @@ function shouldSkipPlanningOnlyRetry(params: {
     params.attempt.yieldDetected ||
     params.attempt.didSendDeterministicApprovalPrompt ||
     params.attempt.lastToolError ||
-    params.attempt.replayMetadata.hadPotentialSideEffects,
+    resolveAttemptReplayMetadata(params.attempt).hadPotentialSideEffects,
   );
 }
 
@@ -796,7 +807,7 @@ export function resolvePlanningOnlyRetryInstruction(params: {
     (hasNonPlanToolActivity(params.attempt.toolMetas) && !allowSingleActionRetryBypass) ||
     (params.attempt.itemLifecycle.startedCount > planOnlyToolMetaCount &&
       !allowSingleActionRetryBypass) ||
-    params.attempt.replayMetadata.hadPotentialSideEffects
+    resolveAttemptReplayMetadata(params.attempt).hadPotentialSideEffects
   ) {
     return null;
   }