diff --git a/src/agents/pi-embedded-helpers.sanitizeuserfacingtext.test.ts b/src/agents/pi-embedded-helpers.sanitizeuserfacingtext.test.ts
index 37603c2627..5b42146114 100644
--- a/src/agents/pi-embedded-helpers.sanitizeuserfacingtext.test.ts
+++ b/src/agents/pi-embedded-helpers.sanitizeuserfacingtext.test.ts
@@ -23,6 +23,29 @@ describe("sanitizeUserFacingText", () => {
     );
   });
 
+  it("sanitizes direct context-overflow errors", () => {
+    expect(
+      sanitizeUserFacingText(
+        "Context overflow: prompt too large for the model. Try again with less input or a larger-context model.",
+      ),
+    ).toContain("Context overflow: prompt too large for the model.");
+    expect(sanitizeUserFacingText("Request size exceeds model context window")).toContain(
+      "Context overflow: prompt too large for the model.",
+    );
+  });
+
+  it("does not rewrite conversational mentions of context overflow", () => {
+    const text =
+      "nah it failed, hit a context overflow. the prompt was too large for the model. want me to retry it with a different approach?";
+    expect(sanitizeUserFacingText(text)).toBe(text);
+  });
+
+  it("does not rewrite technical summaries that mention context overflow", () => {
+    const text =
+      "Problem: When a subagent reads a very large file, it can exceed the model context window. Auto-compaction cannot help in that case.";
+    expect(sanitizeUserFacingText(text)).toBe(text);
+  });
+
   it("sanitizes raw API error payloads", () => {
     const raw = '{"type":"error","error":{"message":"Something exploded","type":"server_error"}}';
     expect(sanitizeUserFacingText(raw)).toBe("LLM error server_error: Something exploded");
diff --git a/src/agents/pi-embedded-helpers/errors.ts b/src/agents/pi-embedded-helpers/errors.ts
index 92a47fd75a..9111020753 100644
--- a/src/agents/pi-embedded-helpers/errors.ts
+++ b/src/agents/pi-embedded-helpers/errors.ts
@@ -67,6 +67,8 @@ const ERROR_PAYLOAD_PREFIX_RE =
 const FINAL_TAG_RE = /<\s*\/?\s*final\s*>/gi;
 const ERROR_PREFIX_RE =
   /^(?:error|api\s*error|openai\s*error|anthropic\s*error|gateway\s*error|request failed|failed|exception)[:\s-]+/i;
+const CONTEXT_OVERFLOW_ERROR_HEAD_RE =
+  /^(?:context overflow:|request_too_large\b|request size exceeds\b|request exceeds the maximum size\b|context length exceeded\b|maximum context length\b|prompt is too long\b|exceeds model context window\b)/i;
 const HTTP_STATUS_PREFIX_RE = /^(?:http\s*)?(\d{3})\s+(.+)$/i;
 const HTTP_ERROR_HINTS = [
   "error",
@@ -135,6 +137,18 @@ function isLikelyHttpErrorText(raw: string): boolean {
   return HTTP_ERROR_HINTS.some((hint) => message.includes(hint));
 }
 
+function shouldRewriteContextOverflowText(raw: string): boolean {
+  if (!isContextOverflowError(raw)) {
+    return false;
+  }
+  return (
+    isRawApiErrorPayload(raw) ||
+    isLikelyHttpErrorText(raw) ||
+    ERROR_PREFIX_RE.test(raw) ||
+    CONTEXT_OVERFLOW_ERROR_HEAD_RE.test(raw)
+  );
+}
+
 type ErrorPayload = Record<string, unknown>;
 
 function isErrorPayloadObject(payload: unknown): payload is ErrorPayload {
@@ -403,7 +417,7 @@ export function sanitizeUserFacingText(text: string): string {
     );
   }
 
-  if (isContextOverflowError(trimmed)) {
+  if (shouldRewriteContextOverflowText(trimmed)) {
     return (
       "Context overflow: prompt too large for the model. " +
       "Try again with less input or a larger-context model."