diff --git a/app/src/components/channels/chat-transcript.tsx b/app/src/components/channels/chat-transcript.tsx
index 56d12b8c0..d6ac4cd2f 100644
--- a/app/src/components/channels/chat-transcript.tsx
+++ b/app/src/components/channels/chat-transcript.tsx
@@ -52,7 +52,12 @@ import { splitChatOrigin } from "./chat-origin";
 import { markdownComponents } from "@/lib/markdown";
 import { EASE_OUT, ENTRANCE_SECONDS } from "@/lib/motion";
 import { readToolName } from "@/lib/plugins/tool-name";
-import { asText, forDisplay, REFUSAL_MARKER } from "@/lib/plugins/tool-result";
+import {
+  asText,
+  forDisplay,
+  REFUSAL_MARKER,
+  toolResultFailed,
+} from "@/lib/plugins/tool-result";
 import { readTeamBotConsent } from "@/lib/team-bots";
 import { cn } from "@/lib/utils";
 import { VOICE_CHAT_ACTIVITY, type VoiceChatEntry } from "@/lib/voice/archive";
@@ -1355,6 +1360,7 @@ function ServerToolLine({ name, result }: { name: string; result?: string }) {
    */
   const answer = result === undefined ? undefined : asText(result);
   const refused = answer?.startsWith(REFUSAL_MARKER) ?? false;
+  const failed = !refused && toolResultFailed(result);
   /*
    * The marker is for this component, not for the reader. Left in, a refusal reads "Blocked" in the
    * label and then "Refused." again in the first two words of the body, which is the same fact three
@@ -1370,6 +1376,7 @@ function ServerToolLine({ name, result }: { name: string; result?: string }) {
       {...(detail ? { detail } : {})}
       label={label}
       refused={refused}
+      failed={failed}
       running={result === undefined}
     >
       {body ? (
diff --git a/app/src/components/channels/tool-line.tsx b/app/src/components/channels/tool-line.tsx
index 643907f1c..1e13bb3c3 100644
--- a/app/src/components/channels/tool-line.tsx
+++ b/app/src/components/channels/tool-line.tsx
@@ -43,7 +43,11 @@ export function ToolLine({
       }`}
     >
       <span className="shrink-0">
-        {refused ? "Blocked" : failed ? `${label}, didn't work` : label}
+        {refused
+          ? `Blocked: ${label}`
+          : failed
+            ? `${label}, didn't work`
+            : label}
       </span>
       {detail ? <span className="truncate opacity-70">{detail}</span> : null}
     </span>
diff --git a/app/src/lib/plugins/tool-result.ts b/app/src/lib/plugins/tool-result.ts
index d6827a512..88f628104 100644
--- a/app/src/lib/plugins/tool-result.ts
+++ b/app/src/lib/plugins/tool-result.ts
@@ -116,3 +116,19 @@ export function saidItWentAhead(result: unknown, marker: string): boolean {
   if (typeof result !== "string") return false;
   return asText(result).startsWith(marker);
 }
+
+/**
+ * Whether a server-side tool reached a terminal failure rather than returning a result.
+ *
+ * These prefixes are written by the two server tool doors after the distinction between policy
+ * refusal and execution failure has already been made. A refusal has its own marker and must not
+ * be collapsed into this state.
+ */
+export function toolResultFailed(result: string | undefined): boolean {
+  if (result === undefined) return false;
+  const answer = asText(result);
+  return (
+    answer.startsWith("The vendor reported an error:") ||
+    answer.startsWith("That tool could not be called")
+  );
+}
diff --git a/app/tests/tool-result.test.ts b/app/tests/tool-result.test.ts
index a700f2064..e2e9b0eec 100644
--- a/app/tests/tool-result.test.ts
+++ b/app/tests/tool-result.test.ts
@@ -4,6 +4,7 @@ import {
   asText,
   forDisplay,
   saidItWentAhead,
+  toolResultFailed,
 } from "../src/lib/plugins/tool-result";
 
 /**
@@ -119,3 +120,24 @@ describe("the markers the server and the transcript both use", () => {
     expect(saidItWentAhead(undefined, HANDED_OVER)).toBe(true);
   });
 });
+
+describe("telling a failed tool from a successful one", () => {
+  test("vendor errors are failures, encoded or not", () => {
+    const answer = "The vendor reported an error:\\npermission denied";
+    expect(toolResultFailed(answer)).toBe(true);
+    expect(toolResultFailed(JSON.stringify(answer))).toBe(true);
+  });
+
+  test("runtime/deployment call failures are failures", () => {
+    expect(toolResultFailed("That tool could not be called.")).toBe(true);
+    expect(
+      toolResultFailed("That tool could not be called: connection reset"),
+    ).toBe(true);
+  });
+
+  test("ordinary results, refusals, and running calls are not failures", () => {
+    expect(toolResultFailed("Found 12 results.")).toBe(false);
+    expect(toolResultFailed("Refused. The rule says no.")).toBe(false);
+    expect(toolResultFailed(undefined)).toBe(false);
+  });
+});