test: stabilize remaining CI flakes

steipete · steipete · commit 526925c509bc · 2026-05-29T23:17:36.000+01:00
diff --git a/src/agents/embedded-agent-runner/run/attempt.cwd-split.test.ts b/src/agents/embedded-agent-runner/run/attempt.cwd-split.test.ts
@@ -1,3 +1,6 @@
+import fs from "node:fs/promises";
+import os from "node:os";
+import path from "node:path";
 import { afterEach, beforeEach, describe, expect, it } from "vitest";
 import {
   cleanupTempPaths,
@@ -22,13 +25,15 @@ describe("runEmbeddedAttempt cwd/workspace split", () => {
 
   it("uses workspace for bootstrap and cwd for runtime tools", async () => {
     const bootstrap = createContextEngineBootstrapAndAssemble();
+    const taskRepo = await fs.mkdtemp(path.join(os.tmpdir(), "openclaw-task-repo-"));
+    tempPaths.push(taskRepo);
 
     await createContextEngineAttemptRunner({
       contextEngine: bootstrap,
       sessionKey: "agent:main:subagent:child",
       tempPaths,
       attemptOverrides: {
-        cwd: "/tmp/task-repo",
+        cwd: taskRepo,
         disableTools: false,
       },
     });
@@ -42,14 +47,14 @@ describe("runEmbeddedAttempt cwd/workspace split", () => {
     const toolsCall = hoisted.createOpenClawCodingToolsMock.mock.calls[0]?.[0] as
       | { cwd?: string; workspaceDir?: string; spawnWorkspaceDir?: string }
       | undefined;
-    expect(toolsCall?.cwd).toBe("/tmp/task-repo");
+    expect(toolsCall?.cwd).toBe(taskRepo);
     expect(toolsCall?.workspaceDir).toBe(bootstrapCall?.workspaceDir);
     expect(toolsCall?.spawnWorkspaceDir).toBe(bootstrapCall?.workspaceDir);
 
     const resourceLoaderInit = hoisted.defaultResourceLoaderInitMock.mock.calls[0]?.[0] as
       | { cwd?: string }
       | undefined;
-    expect(resourceLoaderInit?.cwd).toBe("/tmp/task-repo");
+    expect(resourceLoaderInit?.cwd).toBe(taskRepo);
   });
 
   it("rejects cwd overrides for sandboxed runs instead of silently ignoring them", async () => {
diff --git a/src/gateway/gateway-cli-backend.live-helpers.test.ts b/src/gateway/gateway-cli-backend.live-helpers.test.ts
@@ -23,6 +23,13 @@ vi.mock("./client.js", () => ({
   },
 }));
 
+vi.mock("./client-start-readiness.js", () => ({
+  startGatewayClientWhenEventLoopReady: async (client: { start: () => void }) => {
+    client.start();
+    return { ready: true, aborted: false, elapsedMs: 0, maxDriftMs: 0, checks: 0 };
+  },
+}));
+
 describe("gateway cli backend live helpers", () => {
   let liveHelpers: typeof import("./gateway-cli-backend.live-helpers.js");
 
diff --git a/src/infra/exec-wrapper-trust-plan.test.ts b/src/infra/exec-wrapper-trust-plan.test.ts
@@ -30,16 +30,17 @@ describe("resolveExecWrapperTrustPlan", () => {
       },
     },
     {
-      name: "unwraps script wrappers before evaluating nested shell payloads",
+      name: "blocks script wrappers before evaluating nested shell payloads",
       enabled: process.platform === "darwin" || process.platform === "freebsd",
       argv: ["/usr/bin/script", "-q", "/dev/null", "sh", "-c", "echo hi"],
       expected: {
-        argv: ["sh", "-c", "echo hi"],
-        policyArgv: ["sh", "-c", "echo hi"],
-        wrapperChain: ["script"],
-        policyBlocked: false,
-        shellWrapperExecutable: true,
-        shellInlineCommand: "echo hi",
+        argv: ["/usr/bin/script", "-q", "/dev/null", "sh", "-c", "echo hi"],
+        policyArgv: ["/usr/bin/script", "-q", "/dev/null", "sh", "-c", "echo hi"],
+        wrapperChain: [],
+        policyBlocked: true,
+        blockedWrapper: "script",
+        shellWrapperExecutable: false,
+        shellInlineCommand: null,
       },
     },
     {
