fix(telegram): treat exec exit code errors as internal

Command exit codes (like grep returning 1 for no matches) should not be
forwarded to users. The model sees these errors in its context and can
decide how to proceed - surfacing them clutters the chat with technical
messages that aren't actionable by users.

Add "exited with code", "exit code", and "aborted" to the list of
recoverable/internal error patterns that are suppressed from user output.

Before: "⚠️ 🛠️ Exec: grep ... failed: Command exited with code 1"
After: (error stays internal, model continues normally)

Fixes #3765

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

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

@@ -148,7 +148,7 @@ describe("buildEmbeddedRunPayloads", () => {
     expect(payloads[0]?.text).toBe("All good");
   });
 
-  it("adds tool error fallback when the assistant only invoked tools", () => {
+  it("suppresses exit code errors as internal (model sees them in context)", () => {
     const payloads = buildEmbeddedRunPayloads({
       assistantTexts: [],
       toolMetas: [],
@@ -171,10 +171,9 @@ describe("buildEmbeddedRunPayloads", () => {
       toolResultFormat: "plain",
     });
 
-    expect(payloads).toHaveLength(1);
-    expect(payloads[0]?.isError).toBe(true);
-    expect(payloads[0]?.text).toContain("Exec");
-    expect(payloads[0]?.text).toContain("code 1");
+    // Exit code errors should be treated as internal - the model sees them in context
+    // and can decide how to proceed. Users shouldn't see "grep returned exit code 1" etc.
+    expect(payloads).toHaveLength(0);
   });
 
   it("suppresses recoverable tool errors containing 'required'", () => {
@@ -226,6 +225,22 @@ describe("buildEmbeddedRunPayloads", () => {
     expect(payloads).toHaveLength(0);
   });
 
+  it("suppresses aborted command errors", () => {
+    const payloads = buildEmbeddedRunPayloads({
+      assistantTexts: [],
+      toolMetas: [],
+      lastAssistant: undefined,
+      lastToolError: { toolName: "exec", error: "Command aborted by signal SIGTERM" },
+      sessionKey: "session:telegram",
+      inlineToolResultsAllowed: false,
+      verboseLevel: "off",
+      reasoningLevel: "off",
+      toolResultFormat: "plain",
+    });
+
+    expect(payloads).toHaveLength(0);
+  });
+
   it("shows non-recoverable tool errors to the user", () => {
     const payloads = buildEmbeddedRunPayloads({
       assistantTexts: [],

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

@@ -189,7 +189,12 @@ export function buildEmbeddedRunPayloads(params: {
       errorLower.includes("must be") ||
       errorLower.includes("must have") ||
       errorLower.includes("needs") ||
-      errorLower.includes("requires");
+      errorLower.includes("requires") ||
+      // Treat command exit codes as internal - the model sees them and can decide how to proceed.
+      // Exit code 1 from grep/find/etc. just means "no matches", not a real error.
+      errorLower.includes("exited with code") ||
+      errorLower.includes("exit code") ||
+      errorLower.includes("aborted");
 
     // Show tool errors only when:
     // 1. There's no user-facing reply AND the error is not recoverable