fix(agent): harden runner delivery validation

skills/test/lbs-cli.test.ts

@@ -60,6 +60,7 @@ import {
 } from "../scripts/e2e/lib/debug-chat.mjs";
 import {
   beginBackendLogCapture,
+  clickFirstVisible,
   ensureAuthenticatedBrowser,
   finishBackendLogCapture,
   resolveLangBotRepo,
@@ -68,6 +69,32 @@ import {
 
 const root = process.cwd();
 
+test("clickFirstVisible waits for a later visible DOM match", async () => {
+  let pollCount = 0;
+  let clickedIndex = -1;
+  const emptyLocator = {
+    count: async () => 0,
+    nth: () => { throw new Error("empty locator has no children"); },
+  };
+  const textLocator = {
+    count: async () => 2,
+    nth: (index: number) => ({
+      isVisible: async () => index === 1 && pollCount >= 1,
+      click: async () => { clickedIndex = index; },
+    }),
+  };
+  const page = {
+    getByRole: () => emptyLocator,
+    getByText: () => textLocator,
+    waitForTimeout: async () => { pollCount += 1; },
+  };
+
+  const clicked = await clickFirstVisible(page, ["Debug Chat"], 1_000);
+
+  assert.equal(clicked, "Debug Chat");
+  assert.equal(clickedIndex, 1);
+});
+
 test("repo root detects the skills tree before generated bin exists", () => {
   const tmp = mkdtempSync(join(tmpdir(), "lbs-root-no-bin-"));
   try {
@@ -2064,6 +2091,25 @@ test("debug chat classifier passes when expected text appears in a new assistant
   assert.equal(result.new_assistant_message_count, 1);
 });
 
+test("debug chat classifier rejects a matching assistant message that is not final", () => {
+  const result = classifyDebugChatResult({
+    beforeText: "",
+    afterText: "Bot: E2E_OK:skill",
+    expectedText: "E2E_OK:skill",
+    prompt: "Run the task",
+    latestExpectedLeaf: "E2E_OK:skill",
+    latestFailureLeaf: "",
+    beforeMessages: [],
+    afterMessages: [{ role: "assistant", text: "E2E_OK:skill" }],
+    latestAssistantText: "E2E_OK:skill",
+    latestAssistantIsFinal: false,
+  });
+
+  assert.equal(result.status, "fail");
+  assert.match(result.reason, /was not final/);
+  assert.equal(result.latest_assistant_is_final, false);
+});
+
 test("debug chat classifier accepts formatted responses containing every required fragment", () => {
   const expectedTexts = ["MULTITOOL_COMBO_FINAL", "passcode-6718", "rag-7421", "tool-a", "tool-b"];
   const result = classifyDebugChatResult({

skills/test/test_agent_run_ledger_policy.py

@@ -9,6 +9,8 @@ sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "scripts" / "e2e"))
 
 from agent_run_ledger_policy import (  # noqa: E402
     classify_invalid_tool_argument_errors,
+    classify_tool_authorization,
+    invalid_tool_argument_error_signal,
     load_ledger_json,
 )
 
@@ -64,6 +66,34 @@ class AgentRunLedgerPolicyTests(unittest.TestCase):
         self.assertEqual(failures[0]["kind"], "invalid_json")
         self.assertEqual(failures[0]["field"], "agent_run_event[10].data_json")
 
+    def test_detects_pydantic_tool_argument_validation_error(self) -> None:
+        signal = invalid_tool_argument_error_signal(
+            "1 validation error for RetrieveKnowledgeArgs\n"
+            "kb_id\n  Field required [type=missing, input_value={'top_k': 3}, input_type=dict]"
+        )
+
+        self.assertEqual(signal, "1 validation error for RetrieveKnowledgeArgs")
+
+    def test_ignores_unrelated_tool_execution_error(self) -> None:
+        self.assertEqual(invalid_tool_argument_error_signal("Exit code 1"), "")
+
+    def test_strict_mode_rejects_tool_absent_from_host_authorization(self) -> None:
+        call = {"tool_call_id": "call-1", "tool_name": "Read"}
+
+        failures, warnings = classify_tool_authorization([call], authorization_mode="strict")
+
+        self.assertEqual(failures, [{"kind": "unauthorized_tool_calls", "calls": [call]}])
+        self.assertEqual(warnings, [])
+
+    def test_runner_native_mode_records_non_host_tool_as_warning(self) -> None:
+        call = {"tool_call_id": "call-1", "tool_name": "Read"}
+
+        failures, warnings = classify_tool_authorization([call], authorization_mode="runner-native")
+
+        self.assertEqual(failures, [])
+        self.assertEqual(warnings[0]["kind"], "runner_native_tool_calls")
+        self.assertEqual(warnings[0]["calls"], [call])
+
 
 if __name__ == "__main__":
     unittest.main()