refactor(agents): dedupe transient error copy (#16324)

src/agents/pi-embedded-helpers/errors.ts

@@ -14,6 +14,18 @@ export function formatBillingErrorMessage(provider?: string): string {
 export const BILLING_ERROR_USER_MESSAGE = formatBillingErrorMessage();
 
 const RATE_LIMIT_ERROR_USER_MESSAGE = "⚠️ API rate limit reached. Please try again later.";
+const OVERLOADED_ERROR_USER_MESSAGE =
+  "The AI service is temporarily overloaded. Please try again in a moment.";
+
+function formatRateLimitOrOverloadedErrorCopy(raw: string): string | undefined {
+  if (isRateLimitErrorMessage(raw)) {
+    return RATE_LIMIT_ERROR_USER_MESSAGE;
+  }
+  if (isOverloadedErrorMessage(raw)) {
+    return OVERLOADED_ERROR_USER_MESSAGE;
+  }
+  return undefined;
+}
 
 export function isContextOverflowError(errorMessage?: string): boolean {
   if (!errorMessage) {
@@ -463,12 +475,9 @@ export function formatAssistantErrorText(
     return `LLM request rejected: ${invalidRequest[1]}`;
   }
 
-  if (isRateLimitErrorMessage(raw)) {
+  const transientCopy = formatRateLimitOrOverloadedErrorCopy(raw);
-    return RATE_LIMIT_ERROR_USER_MESSAGE;
+  if (transientCopy) {
-  }
+    return transientCopy;
-
-  if (isOverloadedErrorMessage(raw)) {
-    return "The AI service is temporarily overloaded. Please try again in a moment.";
   }
 
   if (isBillingErrorMessage(raw)) {
@@ -523,11 +532,9 @@ export function sanitizeUserFacingText(text: string, opts?: { errorContext?: boo
     }
 
     if (ERROR_PREFIX_RE.test(trimmed)) {
-      if (isRateLimitErrorMessage(trimmed)) {
+      const prefixedCopy = formatRateLimitOrOverloadedErrorCopy(trimmed);
-        return RATE_LIMIT_ERROR_USER_MESSAGE;
+      if (prefixedCopy) {
-      }
+        return prefixedCopy;
-      if (isOverloadedErrorMessage(trimmed)) {
-        return "The AI service is temporarily overloaded. Please try again in a moment.";
       }
       if (isTimeoutErrorMessage(trimmed)) {
         return "LLM request timed out.";

src/agents/pi-embedded-subscribe.handlers.messages.ts

@@ -57,7 +57,7 @@ export function handleMessageUpdate(
     return;
   }
 
-  ctx.state.lastAssistant = msg;
+  ctx.noteLastAssistant(msg);
 
   const assistantEvent = evt.assistantMessageEvent;
   const assistantRecord =
@@ -200,7 +200,7 @@ export function handleMessageEnd(
   }
 
   const assistantMessage = msg;
-  ctx.state.lastAssistant = assistantMessage;
+  ctx.noteLastAssistant(assistantMessage);
   ctx.recordAssistantUsage((assistantMessage as { usage?: unknown }).usage);
   promoteThinkingTagsToBlocks(assistantMessage);
 

src/agents/pi-embedded-subscribe.handlers.types.ts

@@ -72,6 +72,7 @@ export type EmbeddedPiSubscribeContext = {
   blockChunking?: BlockReplyChunking;
   blockChunker: EmbeddedBlockChunker | null;
   hookRunner?: HookRunner;
+  noteLastAssistant: (msg: AgentMessage) => void;
 
   shouldEmitToolResult: () => boolean;
   shouldEmitToolOutput: () => boolean;

src/agents/pi-embedded-subscribe.ts

@@ -1,3 +1,4 @@
+import type { AgentMessage } from "@mariozechner/pi-agent-core";
 import type { InlineCodeState } from "../markdown/code-spans.js";
 import type {
   EmbeddedPiSubscribeContext,
@@ -569,6 +570,12 @@ export function subscribeEmbeddedPiSession(params: SubscribeEmbeddedPiSessionPar
     resetAssistantMessageState(0);
   };
 
+  const noteLastAssistant = (msg: AgentMessage) => {
+    if (msg?.role === "assistant") {
+      state.lastAssistant = msg;
+    }
+  };
+
   const ctx: EmbeddedPiSubscribeContext = {
     params,
     state,
@@ -576,6 +583,7 @@ export function subscribeEmbeddedPiSession(params: SubscribeEmbeddedPiSessionPar
     blockChunking,
     blockChunker,
     hookRunner: params.hookRunner,
+    noteLastAssistant,
     shouldEmitToolResult,
     shouldEmitToolOutput,
     emitToolSummary,