fix(exec): keep framework verb out of backtick-wrapped failure message (#97319)

ly-wang19 · ly-wang19 · commit 5e72dc8a6b6b · 2026-06-28T12:26:50.000+08:00
diff --git a/src/agents/embedded-agent-runner/run/payloads.errors.test.ts b/src/agents/embedded-agent-runner/run/payloads.errors.test.ts
@@ -781,7 +781,7 @@ describe("buildEmbeddedRunPayloads", () => {
     });
 
     expectSinglePayloadSummary(payloads, {
-      text: "⚠️ 🛠️ `show matrix-progress-@room-@alice:matrix-qa.test-!room:matrix-qa.test.txt (workspace)` failed",
+      text: "⚠️ 🛠️ `matrix-progress-@room-@alice:matrix-qa.test-!room:matrix-qa.test.txt (workspace)` failed",
       isError: true,
     });
   });
diff --git a/src/agents/embedded-agent-runner/run/payloads.ts b/src/agents/embedded-agent-runner/run/payloads.ts
@@ -41,6 +41,7 @@ import {
   extractAssistantThinking,
   extractAssistantVisibleText,
 } from "../../embedded-agent-utils.js";
+import { stripLeadingExecDisplayVerb } from "../../tool-display-exec.js";
 import { isExecLikeToolName, type ToolErrorSummary } from "../../tool-error-summary.js";
 import { isLikelyMutatingToolName } from "../../tool-mutation.js";
 
@@ -550,9 +551,18 @@ export function buildEmbeddedRunPayloads(params: {
     // Surface mutating failures unless the assistant explicitly acknowledged the failed action.
     // Otherwise, keep the previous behavior and only surface non-recoverable failures when no reply exists.
     if (warningPolicy.showWarning) {
+      // For exec/bash failures, strip the leading framework verb from the meta so the
+      // backtick-wrapped text is exactly what was run, not `run <command>`. Otherwise the
+      // verb adjacent to the command in the rendered warning is indistinguishable from an
+      // agent-typed `run <command>` invocation (issue #97319).
+      const meta = params.lastToolError.meta;
+      const displayMeta =
+        isExecLikeToolName(params.lastToolError.toolName) && meta
+          ? stripLeadingExecDisplayVerb(meta)
+          : meta;
       const toolSummary = formatToolAggregate(
         params.lastToolError.toolName,
-        params.lastToolError.meta ? [params.lastToolError.meta] : undefined,
+        displayMeta ? [displayMeta] : undefined,
         { markdown: useMarkdown },
       );
       const errorSuffix =
diff --git a/src/agents/tool-display-exec.ts b/src/agents/tool-display-exec.ts
@@ -287,6 +287,55 @@ function summarizeKnownExec(words: string[]): string {
   return /^[A-Za-z0-9._/-]+$/.test(arg) ? `run ${bin} ${arg}` : `run ${bin}`;
 }
 
+// Known display verbs that summarizeKnownExec emits as the first token of an exec label.
+// Used to split the framework action from the literal command subject so progress and
+// failure surfaces do not backtick-wrap the verb together with the command (issue #97319).
+const EXEC_DISPLAY_VERBS: ReadonlySet<string> = new Set([
+  "run",
+  "check",
+  "view",
+  "show",
+  "list",
+  "switch",
+  "create",
+  "pull",
+  "push",
+  "fetch",
+  "merge",
+  "rebase",
+  "stage",
+  "restore",
+  "reset",
+  "stash",
+  "find",
+  "search",
+  "print",
+  "copy",
+  "move",
+  "remove",
+  "install",
+  "start",
+]);
+
+/**
+ * Strips a leading exec display verb (e.g. "run" in "run python3 /path") so the
+ * framework action label is not backtick-wrapped together with the literal command.
+ *
+ * Returns the original meta when no known exec display verb prefix is present, so
+ * raw shell command text passes through unchanged.
+ */
+export function stripLeadingExecDisplayVerb(meta: string): string {
+  const match = meta.match(/^(\S+)\s+(\S.*)$/);
+  if (!match) {
+    return meta;
+  }
+  const verb = match[1];
+  if (!EXEC_DISPLAY_VERBS.has(verb)) {
+    return meta;
+  }
+  return match[2];
+}
+
 function summarizePipeline(stage: string): string {
   const pipeline = splitTopLevelPipes(stage);
   if (pipeline.length > 1) {
diff --git a/src/agents/tool-display.test.ts b/src/agents/tool-display.test.ts
@@ -4,7 +4,7 @@
  */
 import { describe, expect, it } from "vitest";
 import { resolveToolSearchCodeDisplayTarget } from "./tool-display-common.js";
-import { resolveExecDetail } from "./tool-display-exec.js";
+import { resolveExecDetail, stripLeadingExecDisplayVerb } from "./tool-display-exec.js";
 import { formatToolDetail, formatToolSummary, resolveToolDisplay } from "./tool-display.js";
 
 describe("tool display details", () => {
@@ -586,6 +586,34 @@ describe("compactRawCommand middle truncation", () => {
   });
 });
 
+describe("stripLeadingExecDisplayVerb", () => {
+  it("strips a known exec display verb so it stays outside backtick-wrapped command text", () => {
+    expect(stripLeadingExecDisplayVerb("run python3 /path/to/script.py")).toBe(
+      "python3 /path/to/script.py",
+    );
+    expect(stripLeadingExecDisplayVerb("check git status")).toBe("git status");
+    expect(stripLeadingExecDisplayVerb("show some-file.txt (workspace)")).toBe(
+      "some-file.txt (workspace)",
+    );
+    expect(stripLeadingExecDisplayVerb('find files named "x" in /path')).toBe(
+      'files named "x" in /path',
+    );
+  });
+
+  it("returns the meta unchanged when the first token is not a known exec display verb", () => {
+    // `cd` is a real shell builtin, not a summary verb, so a raw `cd ~/dir && ls` meta survives.
+    expect(stripLeadingExecDisplayVerb("cd ~/dir && ls")).toBe("cd ~/dir && ls");
+    expect(stripLeadingExecDisplayVerb("git status")).toBe("git status");
+    expect(stripLeadingExecDisplayVerb("npm install")).toBe("npm install");
+  });
+
+  it("returns the meta unchanged for single-token or empty input", () => {
+    expect(stripLeadingExecDisplayVerb("run")).toBe("run");
+    expect(stripLeadingExecDisplayVerb("")).toBe("");
+    expect(stripLeadingExecDisplayVerb("   ")).toBe("   ");
+  });
+});
+
 describe("coerceDisplayValue middle truncation", () => {
   it("preserves start and end of long string values", () => {
     const longPath =
