fix: centralize Windows exec handling and async migrations (#318)

* Add Windows exec foundation helpers

* server: centralize windows exec handling phase 2

* migrate cli daemon status to execCommand

* refactor: complete phase 4 async exec migrations

* test: cover windows exec helpers

* fix: use beforeEach for context.skip in test setup

packages/cli/src/commands/daemon/runtime-toolchain.ts

@@ -1,5 +1,5 @@
-import { spawnSync } from "node:child_process";
 import { platform } from "node:os";
+import { execCommand } from "@getpaseo/server";
 
 export interface NodePathFromPidResult {
   nodePath: string | null;
@@ -13,63 +13,37 @@ function normalizeError(error: unknown): string {
   return String(error);
 }
 
-function resolveNodePathFromPidUnix(pid: number): NodePathFromPidResult {
+async function resolveNodePathFromPidUnix(pid: number): Promise<NodePathFromPidResult> {
-  const result = spawnSync("ps", ["-o", "comm=", "-p", String(pid)], {
+  try {
-    encoding: "utf8",
+    const { stdout } = await execCommand("ps", ["-o", "comm=", "-p", String(pid)]);
-    stdio: ["ignore", "pipe", "pipe"],
+    const resolved = stdout.trim();
-  });
-
-  if (result.error) {
-    return { nodePath: null, error: `ps failed: ${normalizeError(result.error)}` };
-  }
-
-  if ((result.status ?? 1) !== 0) {
-    const details = result.stderr?.trim();
-    return {
-      nodePath: null,
-      error: details ? `ps failed: ${details}` : `ps exited with code ${result.status ?? 1}`,
-    };
-  }
-
-  const resolved = result.stdout.trim();
     return resolved
       ? { nodePath: resolved }
       : { nodePath: null, error: "ps returned an empty command path" };
+  } catch (error) {
+    return { nodePath: null, error: `ps failed: ${normalizeError(error)}` };
+  }
 }
 
-function runProcessProbe(
+async function runProcessProbe(
   command: string,
   args: string[],
-): {
+): Promise<{
   resolved: string | null;
   error?: string;
-} {
+}> {
-  const result = spawnSync(command, args, {
+  try {
-    encoding: "utf8",
+    const { stdout } = await execCommand(command, args);
-    stdio: ["ignore", "pipe", "pipe"],
+    const resolved = stdout.trim();
-  });
-
-  if (result.error) {
-    return { resolved: null, error: `${command} failed: ${normalizeError(result.error)}` };
-  }
-
-  if ((result.status ?? 1) !== 0) {
-    const details = result.stderr?.trim();
-    return {
-      resolved: null,
-      error: details
-        ? `${command} failed: ${details}`
-        : `${command} exited with code ${result.status ?? 1}`,
-    };
-  }
-
-  const resolved = result.stdout.trim();
     return resolved
       ? { resolved }
       : { resolved: null, error: `${command} returned no executable path` };
+  } catch (error) {
+    return { resolved: null, error: `${command} failed: ${normalizeError(error)}` };
+  }
 }
 
-function resolveNodePathFromPidWindows(pid: number): NodePathFromPidResult {
+async function resolveNodePathFromPidWindows(pid: number): Promise<NodePathFromPidResult> {
   const probes: Array<{
     label: string;
     command: string;
@@ -103,7 +77,7 @@ function resolveNodePathFromPidWindows(pid: number): NodePathFromPidResult {
 
   const errors: string[] = [];
   for (const probe of probes) {
-    const result = runProcessProbe(probe.command, probe.args);
+    const result = await runProcessProbe(probe.command, probe.args);
     if (result.resolved) {
       const resolved = probe.parseValue ? probe.parseValue(result.resolved) : result.resolved;
       if (resolved) {
@@ -123,8 +97,8 @@ function resolveNodePathFromPidWindows(pid: number): NodePathFromPidResult {
   };
 }
 
-export function resolveNodePathFromPid(pid: number): NodePathFromPidResult {
+export async function resolveNodePathFromPid(pid: number): Promise<NodePathFromPidResult> {
   return platform() === "win32"
-    ? resolveNodePathFromPidWindows(pid)
+    ? await resolveNodePathFromPidWindows(pid)
-    : resolveNodePathFromPidUnix(pid);
+    : await resolveNodePathFromPidUnix(pid);
 }

packages/cli/src/commands/daemon/status.ts

@@ -1,10 +1,11 @@
 import type { Command } from "commander";
-import { execFile } from "node:child_process";
 import { createRequire } from "node:module";
-import { promisify } from "node:util";
+import {
-import { getOrCreateServerId, findExecutable, applyProviderEnv } from "@getpaseo/server";
+  getOrCreateServerId,
-
+  findExecutable,
-const execFileAsync = promisify(execFile);
+  applyProviderEnv,
+  execCommand,
+} from "@getpaseo/server";
 import { tryConnectToDaemon } from "../../utils/client.js";
 import type { CommandOptions, ListResult, OutputSchema } from "../../output/index.js";
 import { resolveLocalDaemonState, resolveTcpHostFromListen } from "./local-daemon.js";
@@ -178,11 +179,9 @@ async function checkProviderBinary(
   }
   const env = applyProviderEnv(process.env);
   try {
-    const { stdout } = await execFileAsync(binaryPath, ["--version"], {
+    const { stdout } = await execCommand(binaryPath, ["--version"], {
-      encoding: "utf8",
       timeout: 5000,
       env,
-      windowsHide: true,
     });
     return { path: binaryPath, version: stdout.trim() || null };
   } catch {
@@ -224,7 +223,7 @@ export async function runStatusCommand(
   if (!state.running) {
     daemonNode = "-";
   } else if (state.pidInfo?.pid) {
-    const fromPid = resolveNodePathFromPid(state.pidInfo.pid);
+    const fromPid = await resolveNodePathFromPid(state.pidInfo.pid);
     daemonNode = fromPid.nodePath ?? `unknown (${fromPid.error ?? "could not resolve from PID"})`;
   } else {
     daemonNode = "unknown (no PID available)";

packages/server/scripts/test-mcp-inject.ts

@@ -123,11 +123,13 @@ async function verifyInjectedMcpForProvider(
 }
 
 async function main(): Promise<void> {
-  if (!isProviderAvailable("claude")) {
+  const claudeAvailable = await isProviderAvailable("claude");
+  if (!claudeAvailable) {
     throw new Error(
       "Claude is not available in this environment. Ensure the `claude` binary and credentials are configured.",
     );
   }
+  const codexAvailable = await isProviderAvailable("codex");
 
   const logger = pino({ level: "silent" });
   const rootCwd = await mkdtemp(path.join(os.tmpdir(), "paseo-mcp-inject-real-"));
@@ -136,7 +138,7 @@ async function main(): Promise<void> {
   const daemon = await createTestPaseoDaemon({
     agentClients: {
       claude: new ClaudeAgentClient({ logger }),
-      ...(isProviderAvailable("codex") ? { codex: new CodexAppServerAgentClient(logger) } : {}),
+      ...(codexAvailable ? { codex: new CodexAppServerAgentClient(logger) } : {}),
     },
     logger,
   });
@@ -159,7 +161,7 @@ async function main(): Promise<void> {
     results.push(claudeResult);
     console.log(`[PASS] Claude MCP injection verified for agent ${claudeResult.agentId}`);
 
-    if (isProviderAvailable("codex")) {
+    if (codexAvailable) {
       const codexResult = await verifyInjectedMcpForProvider(client, "codex", codexCwd);
       createdAgentIds.push(codexResult.agentId);
       results.push(codexResult);

packages/server/src/poc-commands/commands-poc.test.ts

@@ -15,13 +15,12 @@
  * This pattern is used in claude-agent.ts listModels().
  */
 
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { query, type SDKUserMessage } from "@anthropic-ai/claude-agent-sdk";
-import { isCommandAvailableSync } from "../utils/executable.js";
+import { isCommandAvailable } from "../utils/executable.js";
 
 const hasClaudeCredentials =
   !!process.env.CLAUDE_CODE_OAUTH_TOKEN || !!process.env.ANTHROPIC_API_KEY;
-const canRunClaudeIntegration = isCommandAvailableSync("claude") && hasClaudeCredentials;
 
 // Pattern from claude-agent.ts listModels():
 // Use an empty async generator when you just need control methods
@@ -30,10 +29,20 @@ function createEmptyPrompt(): AsyncGenerator<SDKUserMessage, void, undefined> {
 }
 
 describe("Claude Agent SDK Commands POC", () => {
+  let canRunClaudeIntegration = false;
+
+  beforeAll(async () => {
+    canRunClaudeIntegration = (await isCommandAvailable("claude")) && hasClaudeCredentials;
+  });
+
+  beforeEach((context) => {
+    if (!canRunClaudeIntegration) {
+      context.skip();
+    }
+  });
+
   describe("supportedCommands() API", () => {
-    test.runIf(canRunClaudeIntegration)(
+    test("should return an array of SlashCommand objects", async () => {
-      "should return an array of SlashCommand objects",
-      async () => {
       // Use the pattern from claude-agent.ts:
       // Create a query with empty prompt generator for control methods
       const emptyPrompt = createEmptyPrompt();
@@ -72,13 +81,9 @@ describe("Claude Agent SDK Commands POC", () => {
           }
         }
       }
-      },
+    }, 30000);
-      30000,
-    );
 
-    test.runIf(canRunClaudeIntegration)(
+    test("should have valid SlashCommand structure for all commands", async () => {
-      "should have valid SlashCommand structure for all commands",
-      async () => {
       const emptyPrompt = createEmptyPrompt();
 
       const claudeQuery = query({
@@ -109,9 +114,7 @@ describe("Claude Agent SDK Commands POC", () => {
       } finally {
         await claudeQuery.return?.();
       }
-      },
+    }, 30000);
-      30000,
-    );
   });
 
   describe("Command Execution", () => {

packages/server/src/server/agent/providers/__tests__/claude-agent.event-stream.integration.test.ts

@@ -10,14 +10,14 @@
  * CLAUDE_CODE_OAUTH_TOKEN or ANTHROPIC_API_KEY in the environment.
  * They are skipped automatically when credentials are unavailable.
  */
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
 import pino from "pino";
 
 import type { AgentSession, AgentStreamEvent } from "../../agent-sdk-types.js";
-import { isCommandAvailableSync } from "../../../../utils/executable.js";
+import { isCommandAvailable } from "../../../../utils/executable.js";
 import { ClaudeAgentClient } from "../claude-agent.js";
 
 // ---------------------------------------------------------------------------
@@ -28,7 +28,6 @@ const logger = pino({ level: "silent" });
 const client = new ClaudeAgentClient({ logger });
 const hasClaudeCredentials =
   !!process.env.CLAUDE_CODE_OAUTH_TOKEN || !!process.env.ANTHROPIC_API_KEY;
-const canRun = isCommandAvailableSync("claude") && hasClaudeCredentials;
 
 function tmpCwd(prefix: string): string {
   return mkdtempSync(path.join(tmpdir(), prefix));
@@ -194,9 +193,19 @@ function assertInvariants(events: AgentStreamEvent[], foregroundTurnIds: string[
 // ---------------------------------------------------------------------------
 
 describe("Agent event stream redesign — integration", () => {
-  test.skipIf(!canRun)(
+  let canRun = false;
-    "Test 1: Basic foreground turn",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = (await isCommandAvailable("claude")) && hasClaudeCredentials;
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("Test 1: Basic foreground turn", async () => {
     const handle = await createSession({ cwdPrefix: "event-stream-basic-" });
 
     try {
@@ -219,13 +228,9 @@ describe("Agent event stream redesign — integration", () => {
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
-  test.skipIf(!canRun)(
+  test("Test 2: No duplicate user_messages — THE BUG", async () => {
-    "Test 2: No duplicate user_messages — THE BUG",
-    async () => {
     const handle = await createSession({ cwdPrefix: "event-stream-dedup-" });
 
     try {
@@ -248,13 +253,9 @@ describe("Agent event stream redesign — integration", () => {
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
-  test.skipIf(!canRun)(
+  test("Test 3: Lifecycle doesn't get stuck in running", async () => {
-    "Test 3: Lifecycle doesn't get stuck in running",
-    async () => {
     const handle = await createSession({ cwdPrefix: "event-stream-lifecycle-" });
 
     try {
@@ -283,13 +284,9 @@ describe("Agent event stream redesign — integration", () => {
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
-  test.skipIf(!canRun)(
+  test("Test 4: Autonomous run", async () => {
-    "Test 4: Autonomous run",
-    async () => {
     const handle = await createSession({ cwdPrefix: "event-stream-autonomous-" });
     const autonomousWakeToken = `AUTONOMOUS_WAKE_${Date.now().toString(36)}`;
 
@@ -328,22 +325,16 @@ describe("Agent event stream redesign — integration", () => {
 
       // Autonomous turn reaches terminal
       expect(
-          afterForeground.find(
+        afterForeground.find((e) => isTerminalEvent(e) && hasTurnId(e) && e.turnId === autoTurnId),
-            (e) => isTerminalEvent(e) && hasTurnId(e) && e.turnId === autoTurnId,
-          ),
       ).toBeDefined();
 
       assertInvariants(events, [fgTurnId]);
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 90_000);
-    90_000,
-  );
 
-  test.skipIf(!canRun)(
+  test("Test 5: Interruption", async () => {
-    "Test 5: Interruption",
-    async () => {
     const handle = await createSession({ cwdPrefix: "event-stream-interrupt-" });
 
     try {
@@ -426,13 +417,9 @@ describe("Agent event stream redesign — integration", () => {
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
-  test.skipIf(!canRun)(
+  test("Test 6: Sequential foreground turns", async () => {
-    "Test 6: Sequential foreground turns",
-    async () => {
     const handle = await createSession({ cwdPrefix: "event-stream-sequential-" });
 
     try {
@@ -463,13 +450,9 @@ describe("Agent event stream redesign — integration", () => {
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 90_000);
-    90_000,
-  );
 
-  test.skipIf(!canRun)(
+  test("Test 7: Fast-fail", async () => {
-    "Test 7: Fast-fail",
-    async () => {
     const handle = await createSession({ cwdPrefix: "event-stream-fast-fail-" });
 
     try {
@@ -491,21 +474,15 @@ describe("Agent event stream redesign — integration", () => {
 
       // No stale turn_started after terminal
       const terminalIdx = events.indexOf(terminal!);
-        expect(events.slice(terminalIdx + 1).filter((e) => e.type === "turn_started").length).toBe(
+      expect(events.slice(terminalIdx + 1).filter((e) => e.type === "turn_started").length).toBe(0);
-          0,
-        );
 
       assertInvariants(events, [turnId]);
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
-  test.skipIf(!canRun)(
+  test("Test 8: User message dedup by text", async () => {
-    "Test 8: User message dedup by text",
-    async () => {
     const handle = await createSession({ cwdPrefix: "event-stream-user-dedup-" });
 
     try {
@@ -519,7 +496,5 @@ describe("Agent event stream redesign — integration", () => {
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 });

packages/server/src/server/agent/providers/claude-agent-commands.real.e2e.test.ts

@@ -1,14 +1,26 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import pino from "pino";
 
 import type { AgentSlashCommand } from "../agent-sdk-types.js";
-import { isCommandAvailableSync } from "../../../utils/executable.js";
+import { isCommandAvailable } from "../../../utils/executable.js";
 import { ClaudeAgentClient } from "./claude-agent.js";
 
 // Real-Claude contract coverage: validates slash command shape from a live Claude CLI session.
 describe("claude agent commands contract (real)", () => {
+  let canRun = false;
+
+  beforeAll(async () => {
+    canRun = await isCommandAvailable("claude");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
   test("lists slash commands with the expected contract", async () => {
-    expect(isCommandAvailableSync("claude")).toBe(true);
+    expect(await isCommandAvailable("claude")).toBe(true);
 
     const client = new ClaudeAgentClient({
       logger: pino({ level: "silent" }),

packages/server/src/server/agent/providers/claude-agent.integration.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test, beforeAll } from "vitest";
+import { describe, expect, test, beforeAll, beforeEach } from "vitest";
 import { mkdtempSync, readFileSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -6,7 +6,7 @@ import pino from "pino";
 import { query, type SDKUserMessage } from "@anthropic-ai/claude-agent-sdk";
 
 import type { AgentSession, AgentStreamEvent, ToolCallTimelineItem } from "../agent-sdk-types.js";
-import { isCommandAvailableSync } from "../../../utils/executable.js";
+import { isCommandAvailable } from "../../../utils/executable.js";
 import { ClaudeAgentClient } from "./claude-agent.js";
 import { streamSession } from "./test-utils/session-stream-adapter.js";
 
@@ -174,17 +174,22 @@ async function cleanupSession(handle: { cwd: string; session: AgentSession }): P
 }
 
 describe("ClaudeAgentSession integration", () => {
-  const canRunClaudeIntegration = isCommandAvailableSync("claude") && hasClaudeCredentials;
+  let canRunClaudeIntegration = false;
 
-  beforeAll(() => {
+  beforeAll(async () => {
+    canRunClaudeIntegration = (await isCommandAvailable("claude")) && hasClaudeCredentials;
     if (canRunClaudeIntegration) {
-      expect(isCommandAvailableSync("claude")).toBe(true);
+      expect(await isCommandAvailable("claude")).toBe(true);
     }
   });
 
-  test.runIf(canRunClaudeIntegration)(
+  beforeEach((context) => {
-    "streams a basic response turn end-to-end",
+    if (!canRunClaudeIntegration) {
-    async () => {
+      context.skip();
+    }
+  });
+
+  test("streams a basic response turn end-to-end", async () => {
     const handle = await createSession({
       cwdPrefix: "claude-agent-basic-response-",
     });
@@ -213,13 +218,9 @@ describe("ClaudeAgentSession integration", () => {
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
-  test.runIf(canRunClaudeIntegration)(
+  test("keeps bypassPermissions available after a thinking-option restart", async () => {
-    "keeps bypassPermissions available after a thinking-option restart",
-    async () => {
     const handle = await createSession({
       cwdPrefix: "claude-agent-bypass-restart-",
       modeId: "bypassPermissions",
@@ -232,13 +233,9 @@ describe("ClaudeAgentSession integration", () => {
     } finally {
       await cleanupSession(handle);
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
-  test.runIf(canRunClaudeIntegration)(
+  test("supportedModels returns the current abstract Claude SDK model shape", async () => {
-    "supportedModels returns the current abstract Claude SDK model shape",
-    async () => {
     const claudeQuery = query({
       prompt: createEmptyPrompt(),
       options: {
@@ -276,9 +273,7 @@ describe("ClaudeAgentSession integration", () => {
     } finally {
       await claudeQuery.return?.();
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
   test.runIf(canRunClaudeIntegration)(
     "runs a real Bash tool call and completes it",

packages/server/src/server/agent/providers/claude-agent.max-effort.real.e2e.test.ts

@@ -1,8 +1,8 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import pino from "pino";
 
 import type { AgentStreamEvent, AgentSession } from "../agent-sdk-types.js";
-import { isCommandAvailableSync } from "../../../utils/executable.js";
+import { isCommandAvailable } from "../../../utils/executable.js";
 import { ClaudeAgentClient } from "./claude-agent.js";
 import { streamSession } from "./test-utils/session-stream-adapter.js";
 
@@ -29,9 +29,19 @@ async function collectUntilTerminal(session: AgentSession): Promise<AgentStreamE
 }
 
 describe("Claude max effort availability (real)", () => {
-  test.runIf(isCommandAvailableSync("claude") && hasClaudeCredentials)(
+  let canRun = false;
-    "surfaces the Claude stderr diagnostic when bypassPermissions + max effort is unavailable",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = (await isCommandAvailable("claude")) && hasClaudeCredentials;
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("surfaces the Claude stderr diagnostic when bypassPermissions + max effort is unavailable", async () => {
     const client = new ClaudeAgentClient({
       logger: pino({ level: "silent" }),
     });
@@ -57,7 +67,5 @@ describe("Claude max effort availability (real)", () => {
     } finally {
       await session.close().catch(() => undefined);
     }
-    },
+  }, 30_000);
-    30_000,
-  );
 });

packages/server/src/server/agent/providers/claude-agent.ts

@@ -1,5 +1,4 @@
-import { execFile, type ChildProcessWithoutNullStreams } from "node:child_process";
+import { type ChildProcessWithoutNullStreams } from "node:child_process";
-import { promisify } from "node:util";
 import { randomUUID } from "node:crypto";
 import fs from "node:fs";
 import { promises } from "node:fs";
@@ -73,12 +72,11 @@ import type {
   PersistedAgentDescriptor,
 } from "../agent-sdk-types.js";
 import { applyProviderEnv, type ProviderRuntimeSettings } from "../provider-launch-config.js";
-import { findExecutable } from "../../../utils/executable.js";
+import { executableExists, findExecutable } from "../../../utils/executable.js";
-import { spawnProcess } from "../../../utils/spawn.js";
+import { execCommand, spawnProcess } from "../../../utils/spawn.js";
 import { getOrchestratorModeInstructions } from "../orchestrator-instructions.js";
 
 const fsPromises = promises;
-const execFileAsync = promisify(execFile);
 const CLAUDE_SETTING_SOURCES: NonNullable<Options["settingSources"]> = ["user", "project"];
 
 type TurnState = "idle" | "foreground" | "autonomous";
@@ -1130,7 +1128,7 @@ export class ClaudeAgentClient implements AgentClient {
   async isAvailable(): Promise<boolean> {
     const command = this.runtimeSettings?.command;
     if (command?.mode === "replace") {
-      return fs.existsSync(command.argv[0]);
+      return executableExists(command.argv[0]) !== null;
     }
     return true;
   }
@@ -1186,14 +1184,10 @@ async function resolveClaudeVersion(
 
   try {
     if (command?.mode === "replace") {
-      const { stdout } = await execFileAsync(
+      const { stdout } = await execCommand(
         command.argv[0]!,
         [...command.argv.slice(1), "--version"],
-        {
+        { timeout: 5_000 },
-          encoding: "utf8",
-          timeout: 5_000,
-          windowsHide: true,
-        },
       );
       return stdout.trim() || null;
     }
@@ -1203,11 +1197,7 @@ async function resolveClaudeVersion(
       return null;
     }
 
-    const { stdout } = await execFileAsync(executable, ["--version"], {
+    const { stdout } = await execCommand(executable, ["--version"], { timeout: 5_000 });
-      encoding: "utf8",
-      timeout: 5_000,
-      windowsHide: true,
-    });
     return stdout.trim() || null;
   } catch {
     return null;

packages/server/src/server/agent/providers/claude-sdk-behavior.test.ts

@@ -4,9 +4,9 @@
 import { mkdtempSync, rmSync, realpathSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
-import { beforeAll, describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { query, type SDKMessage, type SDKUserMessage } from "@anthropic-ai/claude-agent-sdk";
-import { findExecutable, isCommandAvailableSync } from "../../../utils/executable.js";
+import { findExecutable, isCommandAvailable } from "../../../utils/executable.js";
 
 class Pushable<T> implements AsyncIterable<T> {
   private queue: T[] = [];
@@ -57,17 +57,22 @@ const hasClaudeCredentials =
   !!process.env.CLAUDE_CODE_OAUTH_TOKEN || !!process.env.ANTHROPIC_API_KEY;
 
 describe("Claude SDK direct behavior", () => {
-  const canRunClaudeIntegration = isCommandAvailableSync("claude") && hasClaudeCredentials;
+  let canRunClaudeIntegration = false;
 
-  beforeAll(() => {
+  beforeAll(async () => {
+    canRunClaudeIntegration = (await isCommandAvailable("claude")) && hasClaudeCredentials;
     if (canRunClaudeIntegration) {
-      expect(isCommandAvailableSync("claude")).toBe(true);
+      expect(await isCommandAvailable("claude")).toBe(true);
     }
   });
 
-  test.runIf(canRunClaudeIntegration)(
+  beforeEach((context) => {
-    "shows what happens after interrupt()",
+    if (!canRunClaudeIntegration) {
-    async () => {
+      context.skip();
+    }
+  });
+
+  test("shows what happens after interrupt()", async () => {
     const cwd = tmpCwd();
     const input = new Pushable<SDKUserMessage>();
     const claudeBinary = await findExecutable("claude");
@@ -150,7 +155,5 @@ describe("Claude SDK direct behavior", () => {
       input.end();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 120000);
-    120000,
-  );
 });

packages/server/src/server/agent/providers/codex-app-server-agent.ts

@@ -31,7 +31,7 @@ import type { Logger } from "pino";
 import { execSync } from "node:child_process";
 import type { ChildProcessWithoutNullStreams } from "node:child_process";
 import { randomUUID } from "node:crypto";
-import { existsSync, Dirent } from "node:fs";
+import { Dirent } from "node:fs";
 import fs from "node:fs/promises";
 import os from "node:os";
 import path from "node:path";
@@ -47,7 +47,7 @@ import {
   resolveProviderCommandPrefix,
   type ProviderRuntimeSettings,
 } from "../provider-launch-config.js";
-import { findExecutable } from "../../../utils/executable.js";
+import { executableExists, findExecutable } from "../../../utils/executable.js";
 import { spawnProcess } from "../../../utils/spawn.js";
 import { extractCodexTerminalSessionId, nonEmptyString } from "./tool-call-mapper-utils.js";
 import { buildCodexFeatures, codexModelSupportsFastMode } from "./codex-feature-definitions.js";
@@ -4203,7 +4203,7 @@ export class CodexAppServerAgentClient implements AgentClient {
   async isAvailable(): Promise<boolean> {
     const command = this.runtimeSettings?.command;
     if (command?.mode === "replace") {
-      return existsSync(command.argv[0]);
+      return executableExists(command.argv[0]) !== null;
     }
     return true;
   }

packages/server/src/server/agent/providers/codex-plan-mode.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -12,9 +12,19 @@ function tmpCwd(): string {
 }
 
 describe("Codex app-server provider (real) plan mode", () => {
-  test.runIf(isProviderAvailable("codex"))(
+  let canRun = false;
-    "maps gpt-5.4 markdown plans to a plan tool call instead of todo items",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("codex");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("maps gpt-5.4 markdown plans to a plan tool call instead of todo items", async () => {
     const cwd = tmpCwd();
     const client = new CodexAppServerAgentClient(createTestLogger());
 
@@ -58,7 +68,5 @@ describe("Codex app-server provider (real) plan mode", () => {
     } finally {
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 240_000);
-    240_000,
-  );
 });

packages/server/src/server/agent/providers/diagnostic-utils.ts

@@ -1,9 +1,5 @@
-import { execFile } from "node:child_process";
-import { promisify } from "node:util";
-
 import type { ProviderRuntimeSettings } from "../provider-launch-config.js";
-
+import { execCommand } from "../../../utils/spawn.js";
-const execFileAsync = promisify(execFile);
 
 type DiagnosticEntry = {
   label: string;
@@ -49,11 +45,7 @@ export function toDiagnosticErrorMessage(error: unknown): string {
 
 export async function resolveBinaryVersion(binaryPath: string): Promise<string> {
   try {
-    const { stdout } = await execFileAsync(binaryPath, ["--version"], {
+    const { stdout } = await execCommand(binaryPath, ["--version"], { timeout: 5_000 });
-      encoding: "utf8",
-      timeout: 5_000,
-      windowsHide: true,
-    });
     return stdout.trim() || "unknown";
   } catch {
     return "unknown";

packages/server/src/server/agent/providers/opencode-agent-commands.real.e2e.test.ts

@@ -1,13 +1,25 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import pino from "pino";
 
 import type { AgentSlashCommand } from "../agent-sdk-types.js";
-import { isCommandAvailableSync } from "../../../utils/executable.js";
+import { isCommandAvailable } from "../../../utils/executable.js";
 import { OpenCodeAgentClient } from "./opencode-agent.js";
 
 describe("opencode agent commands contract (real)", () => {
+  let canRun = false;
+
+  beforeAll(async () => {
+    canRun = await isCommandAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
   test("lists slash commands with the expected contract", async () => {
-    expect(isCommandAvailableSync("opencode")).toBe(true);
+    expect(await isCommandAvailable("opencode")).toBe(true);
 
     const client = new OpenCodeAgentClient(pino({ level: "silent" }));
     const session = await client.createSession({
@@ -37,7 +49,7 @@ describe("opencode agent commands contract (real)", () => {
   }, 60_000);
 
   test("executes a slash command without arguments", async () => {
-    expect(isCommandAvailableSync("opencode")).toBe(true);
+    expect(await isCommandAvailable("opencode")).toBe(true);
 
     const client = new OpenCodeAgentClient(pino({ level: "silent" }));
     const session = await client.createSession({

packages/server/src/server/agent/providers/opencode-agent.error-handling.real.e2e.test.ts

@@ -1,8 +1,8 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import pino from "pino";
 
 import type { AgentStreamEvent } from "../agent-sdk-types.js";
-import { isCommandAvailableSync } from "../../../utils/executable.js";
+import { isCommandAvailable } from "../../../utils/executable.js";
 import { OpenCodeAgentClient } from "./opencode-agent.js";
 import { streamSession } from "./test-utils/session-stream-adapter.js";
 
@@ -21,9 +21,19 @@ function isTerminalEvent(event: AgentStreamEvent): boolean {
  * when models are invalid, auth fails, or provider API calls fail.
  */
 describe("opencode agent error handling (real)", () => {
-  test.runIf(isCommandAvailableSync("opencode"))(
+  let canRun = false;
-    "surfaces error for the exact opencode/<bad-model> path used by paseo run",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isCommandAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("surfaces error for the exact opencode/<bad-model> path used by paseo run", async () => {
     const client = new OpenCodeAgentClient(pino({ level: "silent" }));
     const session = await client.createSession({
       provider: "opencode",
@@ -46,13 +56,9 @@ describe("opencode agent error handling (real)", () => {
     } finally {
       await session.close().catch(() => undefined);
     }
-    },
+  }, 45_000);
-    45_000,
-  );
 
-  test.runIf(isCommandAvailableSync("opencode"))(
+  test("surfaces error for unknown provider model (fast path)", async () => {
-    "surfaces error for unknown provider model (fast path)",
-    async () => {
     const client = new OpenCodeAgentClient(pino({ level: "silent" }));
     const session = await client.createSession({
       provider: "opencode",
@@ -75,13 +81,9 @@ describe("opencode agent error handling (real)", () => {
     } finally {
       await session.close().catch(() => undefined);
     }
-    },
+  }, 30_000);
-    30_000,
-  );
 
-  test.runIf(isCommandAvailableSync("opencode"))(
+  test("sequential sessions: second session works after first errors", async () => {
-    "sequential sessions: second session works after first errors",
-    async () => {
     const client = new OpenCodeAgentClient(pino({ level: "silent" }));
 
     // Session 1: bogus model, will error quickly
@@ -114,13 +116,9 @@ describe("opencode agent error handling (real)", () => {
     const terminal = events.find(isTerminalEvent);
     expect(terminal).toBeDefined();
     expect(terminal!.type).toBe("turn_failed");
-    },
+  }, 30_000);
-    30_000,
-  );
 
-  test.runIf(isCommandAvailableSync("opencode"))(
+  test("surfaces error for known provider with nonexistent model (retry path)", async () => {
-    "surfaces error for known provider with nonexistent model (retry path)",
-    async () => {
     // When the provider is recognized (anthropic) but the model doesn't exist,
     // OpenCode retries before surfacing the error. This must not hang.
     const client = new OpenCodeAgentClient(pino({ level: "silent" }));
@@ -148,13 +146,9 @@ describe("opencode agent error handling (real)", () => {
     } finally {
       await session.close().catch(() => undefined);
     }
-    },
+  }, 45_000);
-    45_000,
-  );
 
-  test.runIf(isCommandAvailableSync("opencode"))(
+  test("surfaces fatal retry status from zai/glm-5.1 instead of hanging forever", async () => {
-    "surfaces fatal retry status from zai/glm-5.1 instead of hanging forever",
-    async () => {
     const client = new OpenCodeAgentClient(pino({ level: "silent" }));
     const session = await client.createSession({
       provider: "opencode",
@@ -180,7 +174,5 @@ describe("opencode agent error handling (real)", () => {
     } finally {
       await session.close().catch(() => undefined);
     }
-    },
+  }, 45_000);
-    45_000,
-  );
 });

packages/server/src/server/agent/providers/opencode-agent.ts

@@ -1,5 +1,4 @@
 import type { ChildProcess } from "node:child_process";
-import { existsSync } from "node:fs";
 import {
   createOpencodeClient,
   type AssistantMessage as OpenCodeAssistantMessage,
@@ -44,7 +43,7 @@ import {
   resolveProviderCommandPrefix,
   type ProviderRuntimeSettings,
 } from "../provider-launch-config.js";
-import { findExecutable } from "../../../utils/executable.js";
+import { executableExists, findExecutable } from "../../../utils/executable.js";
 import { spawnProcess } from "../../../utils/spawn.js";
 import { mapOpencodeToolCall } from "./opencode/tool-call-mapper.js";
 import {
@@ -917,7 +916,7 @@ export class OpenCodeAgentClient implements AgentClient {
   async isAvailable(): Promise<boolean> {
     const command = this.runtimeSettings?.command;
     if (command?.mode === "replace") {
-      return existsSync(command.argv[0]);
+      return executableExists(command.argv[0]) !== null;
     }
     return true;
   }

packages/server/src/server/agent/providers/opencode-assistant-message.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, test, expect } from "vitest";
+import { beforeAll, beforeEach, describe, test, expect } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -9,9 +9,19 @@ import { isProviderAvailable } from "../../daemon-e2e/agent-configs.js";
 import type { AgentStreamEvent } from "../agent-sdk-types.js";
 
 describe("OpenCode assistant message", () => {
-  test.runIf(isProviderAvailable("opencode"))(
+  let canRun = false;
-    "assistant_message appears in live stream with opencode/big-pickle",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("assistant_message appears in live stream with opencode/big-pickle", async () => {
     const cwd = mkdtempSync(path.join(tmpdir(), "opencode-msg-"));
     const logger = pino({ level: "silent" });
     const client = new OpenCodeAgentClient(logger);
@@ -32,13 +42,9 @@ describe("OpenCode assistant message", () => {
     } finally {
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 
-  test.runIf(isProviderAvailable("opencode"))(
+  test("streamHistory returns assistant_message after a completed turn", async () => {
-    "streamHistory returns assistant_message after a completed turn",
-    async () => {
     const cwd = mkdtempSync(path.join(tmpdir(), "opencode-history-"));
     const logger = pino({ level: "silent" });
     const client = new OpenCodeAgentClient(logger);
@@ -64,7 +70,5 @@ describe("OpenCode assistant message", () => {
     } finally {
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 });

packages/server/src/server/agent/providers/opencode-reasoning-dedup.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, test, expect } from "vitest";
+import { beforeAll, beforeEach, describe, test, expect } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -9,9 +9,19 @@ import { isProviderAvailable } from "../../daemon-e2e/agent-configs.js";
 import type { AgentStreamEvent } from "../agent-sdk-types.js";
 
 describe("OpenCode reasoning dedup", () => {
-  test.runIf(isProviderAvailable("opencode"))(
+  let canRun = false;
-    "reasoning content is not duplicated as assistant_message",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("reasoning content is not duplicated as assistant_message", async () => {
     const cwd = mkdtempSync(path.join(tmpdir(), "opencode-reasoning-dedup-"));
     const logger = pino({ level: "silent" });
     const client = new OpenCodeAgentClient(logger);
@@ -62,7 +72,5 @@ describe("OpenCode reasoning dedup", () => {
     } finally {
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 120_000);
-    120_000,
-  );
 });

packages/server/src/server/daemon-e2e/agent-configs.ts

@@ -7,9 +7,8 @@ import { join, dirname, resolve } from "node:path";
 import { fileURLToPath } from "node:url";
 import { homedir } from "node:os";
 import dotenv from "dotenv";
-import { isCommandAvailableSync } from "../../utils/executable.js";
+import { isCommandAvailable } from "../../utils/executable.js";
 
-// Load .env.test eagerly so isProviderAvailable() has credentials at collection time.
 const serverRoot = resolve(dirname(fileURLToPath(import.meta.url)), "../../..");
 dotenv.config({ path: resolve(serverRoot, ".env.test"), override: true });
 
@@ -64,6 +63,7 @@ export const agentConfigs = {
 } as const satisfies Record<string, AgentTestConfig>;
 
 export type AgentProvider = keyof typeof agentConfigs;
+const providerAvailabilityCache = new Map<AgentProvider, Promise<boolean>>();
 
 /**
  * Get test config for creating an agent with full permissions (no prompts).
@@ -101,31 +101,42 @@ export function getAskModeConfig(provider: AgentProvider) {
  * This MUST be a function (not a const) so process.env is read at call time,
  * after dotenv has injected the test credentials.
  */
-export function isProviderAvailable(provider: AgentProvider): boolean {
+export function isProviderAvailable(provider: AgentProvider): Promise<boolean> {
+  const cached = providerAvailabilityCache.get(provider);
+  if (cached) {
+    return cached;
+  }
+
+  const availability = (async (): Promise<boolean> => {
     switch (provider) {
       case "claude":
         return (
-        isCommandAvailableSync("claude") &&
+          (await isCommandAvailable("claude")) &&
           (Boolean(process.env.CLAUDE_CODE_OAUTH_TOKEN) || Boolean(process.env.ANTHROPIC_API_KEY))
         );
       case "codex":
         return (
-        isCommandAvailableSync("codex") &&
+          (await isCommandAvailable("codex")) &&
-        (existsSync(join(homedir(), ".codex", "auth.json")) || Boolean(process.env.OPENAI_API_KEY))
+          (existsSync(join(homedir(), ".codex", "auth.json")) ||
+            Boolean(process.env.OPENAI_API_KEY))
         );
       case "copilot":
-      return isCommandAvailableSync("copilot");
+        return await isCommandAvailable("copilot");
       case "opencode":
-      return isCommandAvailableSync("opencode");
+        return await isCommandAvailable("opencode");
       case "pi":
         return (
-        isCommandAvailableSync(process.env.PI_ACP_PI_COMMAND ?? "pi") &&
+          (await isCommandAvailable(process.env.PI_ACP_PI_COMMAND ?? "pi")) &&
           (Boolean(process.env.OPENAI_API_KEY) ||
             Boolean(process.env.ANTHROPIC_API_KEY) ||
             Boolean(process.env.OPENROUTER_API_KEY) ||
             existsSync(join(homedir(), ".pi", "agent", "auth.json")))
         );
     }
+  })();
+
+  providerAvailabilityCache.set(provider, availability);
+  return availability;
 }
 
 /**

packages/server/src/server/daemon-e2e/claude-autonomous-wake-simple.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -18,9 +18,19 @@ function compactText(value: string): string {
 }
 
 describe("daemon E2E (real claude) - autonomous wake simple", () => {
-  test.runIf(isProviderAvailable("claude"))(
+  let canRun = false;
-    "hello + background sleep returns idle, then wakes once on completion",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("claude");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("hello + background sleep returns idle, then wakes once on completion", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -109,7 +119,5 @@ describe("daemon E2E (real claude) - autonomous wake simple", () => {
       await daemon.close().catch(() => undefined);
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 420_000);
-    420_000,
-  );
 });

packages/server/src/server/daemon-e2e/claude-autonomous-wake.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { existsSync, mkdtempSync, readFileSync, realpathSync, rmSync } from "node:fs";
 import { homedir, tmpdir } from "node:os";
 import path from "node:path";
@@ -353,9 +353,19 @@ function summarizeTimelineEntry(entry: {
 }
 
 describe("daemon E2E (real claude) - autonomous wake from background task", () => {
-  test.runIf(isProviderAvailable("claude"))(
+  let canRun = false;
-    "A: background sleep returns idle, then wakes autonomously and appends timeline activity",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("claude");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("A: background sleep returns idle, then wakes autonomously and appends timeline activity", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -427,13 +437,9 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 420_000);
-    420_000,
-  );
 
-  test.runIf(isProviderAvailable("claude"))(
+  test("B: immediate HELLO before task notification returns promptly without deadlock", async () => {
-    "B: immediate HELLO before task notification returns promptly without deadlock",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -485,13 +491,9 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 420_000);
-    420_000,
-  );
 
-  test.runIf(isProviderAvailable("claude"))(
+  test("C: interrupt during overlap returns quickly and does not leave agent stuck running", async () => {
-    "C: interrupt during overlap returns quickly and does not leave agent stuck running",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -555,13 +557,9 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 600_000);
-    600_000,
-  );
 
-  test.runIf(isProviderAvailable("claude"))(
+  test("returns to running after background sleep completes without a second prompt", async () => {
-    "returns to running after background sleep completes without a second prompt",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -617,13 +615,9 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 420_000);
-    420_000,
-  );
 
-  test.runIf(isProviderAvailable("claude"))(
+  test("accepts a new prompt after background sleep finishes and replies HELLO", async () => {
-    "accepts a new prompt after background sleep finishes and replies HELLO",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -666,13 +660,9 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 420_000);
-    420_000,
-  );
 
-  test.runIf(isProviderAvailable("claude"))(
+  test("repro: do-it-again + immediate hello can hang after autonomous wake under churn", async () => {
-    "repro: do-it-again + immediate hello can hang after autonomous wake under churn",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -773,13 +763,9 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 900_000);
-    900_000,
-  );
 
-  test.runIf(isProviderAvailable("claude"))(
+  test("repro: second background sleep completion after HELLO should settle back to idle", async () => {
-    "repro: second background sleep completion after HELLO should settle back to idle",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -849,13 +835,9 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 600_000);
-    600_000,
-  );
 
-  test.runIf(isProviderAvailable("claude"))(
+  test("stress: immediate HELLO before task notification should not leave autonomous run stuck", async () => {
-    "stress: immediate HELLO before task notification should not leave autonomous run stuck",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -939,13 +921,9 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 900_000);
-    900_000,
-  );
 
-  test.runIf(isProviderAvailable("claude"))(
+  test("repro: transcript/timeline parity after do-it-again + hello race (hang + interrupt + drop)", async () => {
-    "repro: transcript/timeline parity after do-it-again + hello race (hang + interrupt + drop)",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -1086,7 +1064,5 @@ describe("daemon E2E (real claude) - autonomous wake from background task", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 600_000);
-    600_000,
-  );
 });

packages/server/src/server/daemon-e2e/model-resolution-on-init.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -14,9 +14,19 @@ function tmpCwd(): string {
 }
 
 describe("daemon E2E (real claude) - model resolution on init", () => {
-  test.runIf(isProviderAvailable("claude"))(
+  let canRun = false;
-    "runtimeInfo.model is set as soon as the agent starts running, not after turn completes",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("claude");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("runtimeInfo.model is set as soon as the agent starts running, not after turn completes", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -59,7 +69,5 @@ describe("daemon E2E (real claude) - model resolution on init", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 120_000);
-    120_000,
-  );
 });

packages/server/src/server/daemon-e2e/model-runtime-reconcile-claude.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -14,9 +14,19 @@ function tmpCwd(): string {
 }
 
 describe("daemon E2E (real claude) - runtime model reconciliation", () => {
-  test.runIf(isProviderAvailable("claude"))(
+  let canRun = false;
-    "normalizes runtime model to a model ID exposed by the provider catalog",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("claude");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("normalizes runtime model to a model ID exposed by the provider catalog", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -57,7 +67,5 @@ describe("daemon E2E (real claude) - runtime model reconciliation", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 300_000);
-    300_000,
-  );
 });

packages/server/src/server/daemon-e2e/opencode-custom-agents.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, test, expect } from "vitest";
+import { beforeAll, beforeEach, describe, test, expect } from "vitest";
 import { mkdtempSync, rmSync, writeFileSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -29,9 +29,19 @@ async function createHarness(): Promise<{
 }
 
 describe("daemon E2E (real opencode) - custom agent discovery", () => {
-  test.runIf(isProviderAvailable("opencode"))(
+  let canRun = false;
-    "custom agents defined in opencode.json appear in availableModes",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("custom agents defined in opencode.json appear in availableModes", async () => {
     const cwd = tmpCwd();
     writeFileSync(
       path.join(cwd, "opencode.json"),
@@ -77,7 +87,5 @@ describe("daemon E2E (real opencode) - custom agent discovery", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 });

packages/server/src/server/daemon-e2e/opencode-initial-prompt-wait.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -29,9 +29,19 @@ async function createHarness(): Promise<{
 }
 
 describe("daemon E2E (real opencode) - initial prompt wait", () => {
-  test.runIf(isProviderAvailable("opencode"))(
+  let canRun = false;
-    "waitForFinish does not resolve before an initial prompt using opencode/big-pickle actually completes",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("waitForFinish does not resolve before an initial prompt using opencode/big-pickle actually completes", async () => {
     const cwd = tmpCwd();
     const { client, daemon } = await createHarness();
 
@@ -72,13 +82,9 @@ describe("daemon E2E (real opencode) - initial prompt wait", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 90_000);
-    90_000,
-  );
 
-  test.runIf(isProviderAvailable("opencode"))(
+  test("waitForFinish surfaces a terminal error when zai/glm-5.1 enters a fatal retry loop", async () => {
-    "waitForFinish surfaces a terminal error when zai/glm-5.1 enters a fatal retry loop",
-    async () => {
     const cwd = tmpCwd();
     const { client, daemon } = await createHarness();
 
@@ -107,7 +113,5 @@ describe("daemon E2E (real opencode) - initial prompt wait", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 90_000);
-    90_000,
-  );
 });

packages/server/src/server/daemon-e2e/opencode-invalid-model.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -29,9 +29,19 @@ async function createHarness(): Promise<{
 }
 
 describe("daemon E2E (real opencode) - invalid model handling", () => {
-  test.runIf(isProviderAvailable("opencode"))(
+  let canRun = false;
-    "initial prompt with a nonexistent OpenCode model fails instead of hanging forever",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("initial prompt with a nonexistent OpenCode model fails instead of hanging forever", async () => {
     const cwd = tmpCwd();
     const { client, daemon } = await createHarness();
 
@@ -55,7 +65,5 @@ describe("daemon E2E (real opencode) - invalid model handling", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 60_000);
-    60_000,
-  );
 });

packages/server/src/server/daemon-e2e/opencode-plan-and-questions.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, test, expect } from "vitest";
+import { beforeAll, beforeEach, describe, test, expect } from "vitest";
 import { existsSync, mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -39,9 +39,19 @@ async function createHarness(): Promise<{
 }
 
 describe("daemon E2E (real opencode) - plan mode and clarifying questions", () => {
-  test.runIf(isProviderAvailable("opencode"))(
+  let canRun = false;
-    "surfaces clarifying questions as pending permissions",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("surfaces clarifying questions as pending permissions", async () => {
     const cwd = tmpCwd();
     const { client, daemon } = await createHarness();
 
@@ -86,13 +96,9 @@ describe("daemon E2E (real opencode) - plan mode and clarifying questions", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 180_000);
-    180_000,
-  );
 
-  test.runIf(isProviderAvailable("opencode"))(
+  test("plan mode stays read-only through the daemon path", async () => {
-    "plan mode stays read-only through the daemon path",
-    async () => {
     const cwd = tmpCwd();
     const filePath = path.join(cwd, "plan-mode-output.txt");
     const { client, daemon } = await createHarness();
@@ -134,7 +140,5 @@ describe("daemon E2E (real opencode) - plan mode and clarifying questions", () =
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 240_000);
-    240_000,
-  );
 });

packages/server/src/server/daemon-e2e/opencode-send-interrupt.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -281,9 +281,19 @@ async function createHarness(): Promise<{
 }
 
 describe("daemon E2E (real opencode) - send while working and interrupt", () => {
-  test.runIf(isProviderAvailable("opencode"))(
+  let canRun = false;
-    "send_message while sleep tool call is running starts a clean replacement turn",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("opencode");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("send_message while sleep tool call is running starts a clean replacement turn", async () => {
     const cwd = tmpCwd();
     const { client, daemon } = await createHarness();
     const collector = createMessageCollector(client);
@@ -341,13 +351,9 @@ describe("daemon E2E (real opencode) - send while working and interrupt", () =>
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 360_000);
-    360_000,
-  );
 
-  test.runIf(isProviderAvailable("opencode"))(
+  test("explicit interrupt during sleep tool call still allows the next turn to complete", async () => {
-    "explicit interrupt during sleep tool call still allows the next turn to complete",
-    async () => {
     const cwd = tmpCwd();
     const { client, daemon } = await createHarness();
     const collector = createMessageCollector(client);
@@ -401,9 +407,9 @@ describe("daemon E2E (real opencode) - send while working and interrupt", () =>
       expect(postInterruptAssistantTexts.some((text) => text.includes("[System Error]"))).toBe(
         false,
       );
-        expect(
+      expect(postInterruptAssistantTexts.some((text) => text.includes(SYSTEM_ERROR_SNIPPET))).toBe(
-          postInterruptAssistantTexts.some((text) => text.includes(SYSTEM_ERROR_SNIPPET)),
+        false,
-        ).toBe(false);
+      );
 
       const timeline = await client.fetchAgentTimeline(agent.id, { limit: 200 });
       const assistantTexts = getTimelineAssistantTexts(timeline);
@@ -416,7 +422,5 @@ describe("daemon E2E (real opencode) - send while working and interrupt", () =>
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 360_000);
-    360_000,
-  );
 });

packages/server/src/server/daemon-e2e/pi.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -16,9 +16,19 @@ function tmpCwd(): string {
 }
 
 describe("daemon E2E (real pi)", () => {
-  test.runIf(isProviderAvailable("pi"))(
+  let canRun = false;
-    "smoke test with thinking option configured separately from modes",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("pi");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("smoke test with thinking option configured separately from modes", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -70,7 +80,5 @@ describe("daemon E2E (real pi)", () => {
       await daemon.close().catch(() => undefined);
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 420_000);
-    420_000,
-  );
 });

packages/server/src/server/daemon-e2e/rewind-user-message-dedupe-claude.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -14,9 +14,19 @@ function tmpCwd(): string {
 }
 
 describe("daemon E2E (real claude) - rewind user message dedupe", () => {
-  test.runIf(isProviderAvailable("claude"))(
+  let canRun = false;
-    "emits /rewind user message once in persisted timeline",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("claude");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("emits /rewind user message once in persisted timeline", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -61,7 +71,5 @@ describe("daemon E2E (real claude) - rewind user message dedupe", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 300_000);
-    300_000,
-  );
 });

packages/server/src/server/daemon-e2e/send-during-tool-call-claude.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, test, expect } from "vitest";
+import { beforeAll, beforeEach, describe, test, expect } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -168,9 +168,19 @@ async function waitForRunningToolCall(
 }
 
 describe("daemon E2E (real claude) - send message during tool call", () => {
-  test.runIf(isProviderAvailable("claude"))(
+  let canRun = false;
-    "sending a message while a tool call is running replaces the turn without error, idle flash, or autonomous fallback",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("claude");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("sending a message while a tool call is running replaces the turn without error, idle flash, or autonomous fallback", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -272,7 +282,5 @@ describe("daemon E2E (real claude) - send message during tool call", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 300_000);
-    300_000,
-  );
 });

packages/server/src/server/daemon-e2e/send-while-running-stuck-claude.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, test, expect } from "vitest";
+import { beforeAll, beforeEach, describe, test, expect } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -15,9 +15,19 @@ function tmpCwd(): string {
 }
 
 describe("daemon E2E (real claude) - send while running recovery", () => {
-  test.runIf(isProviderAvailable("claude"))(
+  let canRun = false;
-    "clears input processing when the interrupt transition is missed",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("claude");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("clears input processing when the interrupt transition is missed", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -117,7 +127,5 @@ describe("daemon E2E (real claude) - send while running recovery", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 300_000);
-    300_000,
-  );
 });

packages/server/src/server/daemon-e2e/send-while-running-stuck.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, test, expect } from "vitest";
+import { beforeAll, beforeEach, describe, test, expect } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -15,9 +15,19 @@ function tmpCwd(): string {
 }
 
 describe("daemon E2E (real codex) - send while running recovery", () => {
-  test.runIf(isProviderAvailable("codex"))(
+  let canRun = false;
-    "clears input processing when the interrupt transition is missed",
+
-    async () => {
+  beforeAll(async () => {
+    canRun = await isProviderAvailable("codex");
+  });
+
+  beforeEach((context) => {
+    if (!canRun) {
+      context.skip();
+    }
+  });
+
+  test("clears input processing when the interrupt transition is missed", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -114,7 +124,5 @@ describe("daemon E2E (real codex) - send while running recovery", () => {
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 240_000);
-    240_000,
-  );
 });

packages/server/src/server/daemon-e2e/ui-action-stress.real.e2e.test.ts

@@ -1,4 +1,4 @@
-import { describe, expect, test } from "vitest";
+import { beforeAll, beforeEach, describe, expect, test } from "vitest";
 import { mkdtempSync, rmSync } from "node:fs";
 import { tmpdir } from "node:os";
 import path from "node:path";
@@ -456,11 +456,19 @@ function createRealAgentClient(provider: AgentProvider, logger: pino.Logger): Ag
 }
 
 describe.each(allProviders)("daemon E2E (real %s) - UI action stress", (provider) => {
-  const shouldRun = isProviderAvailable(provider);
+  let shouldRun = false;
 
-  test.runIf(shouldRun)(
+  beforeAll(async () => {
-    "normal UI submit path (idle sends) stays correct",
+    shouldRun = await isProviderAvailable(provider);
-    async () => {
+  });
+
+  beforeEach((context) => {
+    if (!shouldRun) {
+      context.skip();
+    }
+  });
+
+  test("normal UI submit path (idle sends) stays correct", async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -494,13 +502,9 @@ describe.each(allProviders)("daemon E2E (real %s) - UI action stress", (provider
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 420_000);
-    420_000,
-  );
 
-  test.runIf(shouldRun)(
+  test("queued-send-now path is stable under overlap", async () => {
-    "queued-send-now path is stable under overlap",
-    async () => {
     const logger = pino({ level: "silent" });
     const cwd = tmpCwd();
     const daemon = await createTestPaseoDaemon({
@@ -541,7 +545,5 @@ describe.each(allProviders)("daemon E2E (real %s) - UI action stress", (provider
       await daemon.close();
       rmSync(cwd, { recursive: true, force: true });
     }
-    },
+  }, 600_000);
-    600_000,
-  );
 });

packages/server/src/server/exports.ts

@@ -33,11 +33,11 @@ export {
 export { applyProviderEnv } from "./agent/provider-launch-config.js";
 export {
   findExecutable,
-  findExecutableSync,
+  executableExists,
   quoteWindowsArgument,
   quoteWindowsCommand,
 } from "../utils/executable.js";
-export { spawnProcess } from "../utils/spawn.js";
+export { execCommand, spawnProcess } from "../utils/spawn.js";
 
 // Provider manifest (source of truth for provider definitions)
 export {

packages/server/src/server/loop-service.ts

@@ -1,8 +1,6 @@
 import { randomUUID } from "node:crypto";
-import { execFile } from "node:child_process";
 import { promises as fs } from "node:fs";
 import path from "node:path";
-import { promisify } from "node:util";
 import { z } from "zod";
 import type { Logger } from "pino";
 import { curateAgentActivity } from "./agent/activity-curator.js";
@@ -15,8 +13,8 @@ import type {
   AgentTimelineItem,
   AgentProvider,
 } from "./agent/agent-sdk-types.js";
+import { execCommand, platformShell } from "../utils/spawn.js";
 
-const execFileAsync = promisify(execFile);
 const LOOP_ID_LENGTH = 8;
 const DEFAULT_LOOP_PROVIDER: AgentProvider = "claude";
 const MAX_VERIFY_OUTPUT_BYTES = 64 * 1024;
@@ -256,7 +254,8 @@ async function runVerifyCheck(options: {
 }): Promise<LoopVerifyCheckResult> {
   const startedAt = nowIso();
   try {
-    const result = await execFileAsync("/bin/sh", ["-lc", options.command], {
+    const shell = platformShell();
+    const result = await execCommand(shell.command, [...shell.flag, options.command], {
       cwd: options.cwd,
       maxBuffer: MAX_VERIFY_OUTPUT_BYTES,
     });

packages/server/src/server/session.ts

@@ -1,7 +1,7 @@
 import equal from "fast-deep-equal";
 import { v4 as uuidv4 } from "uuid";
 import { stat } from "fs/promises";
-import { exec, execFile } from "node:child_process";
+import { exec } from "node:child_process";
 import { promisify } from "util";
 import { resolve, sep } from "path";
 import { homedir } from "node:os";
@@ -170,6 +170,7 @@ import { ChatServiceError, FileBackedChatService } from "./chat/chat-service.js"
 import { notifyChatMentions } from "./chat/chat-mentions.js";
 import { LoopService } from "./loop-service.js";
 import { ScheduleService } from "./schedule/service.js";
+import { execCommand } from "../utils/spawn.js";
 import {
   assertSafeGitRef as assertWorktreeSafeGitRef,
   buildAgentSessionConfig as buildWorktreeAgentSessionConfig,
@@ -182,7 +183,6 @@ import {
 } from "./worktree-session.js";
 
 const execAsync = promisify(exec);
-const execFileAsync = promisify(execFile);
 const MAX_INITIAL_AGENT_TITLE_CHARS = Math.min(60, MAX_EXPLICIT_AGENT_TITLE_CHARS);
 const pendingAgentInitializations = new Map<string, Promise<ManagedAgent>>();
 const DEFAULT_AGENT_PROVIDER = AGENT_PROVIDER_IDS[0];
@@ -3531,7 +3531,7 @@ export class Session {
   private async checkoutExistingBranch(cwd: string, branch: string): Promise<void> {
     this.assertSafeGitRef(branch, "branch");
     try {
-      await execFileAsync("git", ["rev-parse", "--verify", branch], { cwd });
+      await execCommand("git", ["rev-parse", "--verify", branch], { cwd });
     } catch (error) {
       throw new Error(`Branch not found: ${branch}`);
     }
@@ -3545,7 +3545,7 @@ export class Session {
     }
 
     await this.ensureCleanWorkingTree(cwd);
-    await execFileAsync("git", ["checkout", branch], { cwd });
+    await execCommand("git", ["checkout", branch], { cwd });
   }
 
   private async createBranchFromBase(params: {
@@ -3558,7 +3558,7 @@ export class Session {
     this.assertSafeGitRef(newBranchName, "new branch");
 
     try {
-      await execFileAsync("git", ["rev-parse", "--verify", baseBranch], { cwd });
+      await execCommand("git", ["rev-parse", "--verify", baseBranch], { cwd });
     } catch (error) {
       throw new Error(`Base branch not found: ${baseBranch}`);
     }
@@ -3569,7 +3569,7 @@ export class Session {
     }
 
     await this.ensureCleanWorkingTree(cwd);
-    await execFileAsync("git", ["checkout", "-b", newBranchName, baseBranch], {
+    await execCommand("git", ["checkout", "-b", newBranchName, baseBranch], {
       cwd,
     });
   }
@@ -3577,7 +3577,7 @@ export class Session {
   private async doesLocalBranchExist(cwd: string, branch: string): Promise<boolean> {
     this.assertSafeGitRef(branch, "branch");
     try {
-      await execFileAsync("git", ["show-ref", "--verify", "--quiet", `refs/heads/${branch}`], {
+      await execCommand("git", ["show-ref", "--verify", "--quiet", `refs/heads/${branch}`], {
         cwd,
       });
       return true;
@@ -4030,7 +4030,7 @@ export class Session {
 
       // Try local branch first
       try {
-        await execFileAsync("git", ["rev-parse", "--verify", branchName], {
+        await execCommand("git", ["rev-parse", "--verify", branchName], {
           cwd: resolvedCwd,
           env: READ_ONLY_GIT_ENV,
         });
@@ -4051,7 +4051,7 @@ export class Session {
 
       // Try remote branch (origin/{branchName})
       try {
-        await execFileAsync("git", ["rev-parse", "--verify", `origin/${branchName}`], {
+        await execCommand("git", ["rev-parse", "--verify", `origin/${branchName}`], {
           cwd: resolvedCwd,
           env: READ_ONLY_GIT_ENV,
         });
@@ -4281,7 +4281,7 @@ export class Session {
       const message = branchLabel
         ? `${Session.PASEO_STASH_PREFIX} ${branchLabel}`
         : `${Session.PASEO_STASH_PREFIX} unnamed`;
-      await execFileAsync("git", ["stash", "push", "--include-untracked", "-m", message], { cwd });
+      await execCommand("git", ["stash", "push", "--include-untracked", "-m", message], { cwd });
       this.checkoutDiffManager.scheduleRefreshForCwd(cwd);
       this.emit({
         type: "stash_save_response",
@@ -4300,7 +4300,7 @@ export class Session {
   ): Promise<void> {
     const { cwd, stashIndex, requestId } = msg;
     try {
-      await execFileAsync("git", ["stash", "pop", `stash@{${stashIndex}}`], { cwd });
+      await execCommand("git", ["stash", "pop", `stash@{${stashIndex}}`], { cwd });
       this.checkoutDiffManager.scheduleRefreshForCwd(cwd);
       this.emit({
         type: "stash_pop_response",

packages/server/src/tasks/cli.ts

@@ -1,11 +1,11 @@
 #!/usr/bin/env node
 import { Command } from "commander";
-import { spawnSync } from "node:child_process";
 import { appendFileSync, existsSync, openSync, readFileSync } from "node:fs";
 import { resolve } from "node:path";
 import { FileTaskStore } from "./task-store.js";
 import { computeExecutionOrder, buildSortedChildrenMap } from "./execution-order.js";
 import { resolvePackageVersion } from "../server/package-version.js";
+import { spawnProcess } from "../utils/spawn.js";
 import type { AgentType, ModelName, Task } from "./types.js";
 
 const TASKS_DIR = resolve(process.cwd(), ".tasks");
@@ -638,11 +638,11 @@ function getAgentConfig(modelStr: string): AgentConfig {
   return { cli: "claude", model };
 }
 
-function runAgentWithModel(
+async function runAgentWithModel(
   prompt: string,
   modelStr: string,
   logFile: string,
-): { success: boolean; output: string } {
+): Promise<{ success: boolean; output: string }> {
   const config = getAgentConfig(modelStr);
   let args: string[];
 
@@ -667,20 +667,34 @@ function runAgentWithModel(
     args.push(prompt);
   }
 
-  const result = spawnSync(config.cli, args, {
+  const { stdout, stderr, exitCode } = await new Promise<{
+    stdout: string;
+    stderr: string;
+    exitCode: number | null;
+  }>((resolve, reject) => {
+    const child = spawnProcess(config.cli, args, {
       stdio: ["inherit", "pipe", "pipe"],
       cwd: process.cwd(),
-    maxBuffer: 50 * 1024 * 1024,
     });
-
+    const stdoutChunks: Buffer[] = [];
-  const stdout = result.stdout?.toString() ?? "";
+    const stderrChunks: Buffer[] = [];
-  const stderr = result.stderr?.toString() ?? "";
+    child.stdout?.on("data", (chunk: Buffer) => stdoutChunks.push(chunk));
+    child.stderr?.on("data", (chunk: Buffer) => stderrChunks.push(chunk));
+    child.on("error", reject);
+    child.on("close", (code) => {
+      resolve({
+        stdout: Buffer.concat(stdoutChunks).toString("utf8"),
+        stderr: Buffer.concat(stderrChunks).toString("utf8"),
+        exitCode: code,
+      });
+    });
+  });
   const output = stdout + stderr;
 
   // Append to log file for history
   appendFileSync(logFile, output);
 
-  return { success: result.status === 0, output };
+  return { success: exitCode === 0, output };
 }
 
 async function buildTaskContext(task: Task, scopeId?: string): Promise<string> {
@@ -1102,7 +1116,7 @@ program
       log(logFile, `[PLANNER] Running ${plannerModel} (${reason})...`);
       const context = await buildTaskContext(task, scopeId);
       const plannerPrompt = makePlannerPrompt(task, context, reason);
-      runAgentWithModel(plannerPrompt, plannerModel, logFile);
+      await runAgentWithModel(plannerPrompt, plannerModel, logFile);
 
       // Check if planner created subtasks for this task
       const children = await store.getChildren(task.id);
@@ -1221,7 +1235,7 @@ program
 
           const freshContext = await buildTaskContext(freshTask, scopeId);
           const workerPrompt = makeWorkerPrompt(freshTask, freshContext, iteration);
-          runAgentWithModel(workerPrompt, workerModel, logFile);
+          await runAgentWithModel(workerPrompt, workerModel, logFile);
 
           // Step 3: Judge
           log(logFile, `[JUDGE] Verifying with ${judgeModel}...`);
@@ -1229,7 +1243,7 @@ program
           if (!judgeTask) break;
 
           const judgePrompt = makeJudgePrompt(judgeTask);
-          const judgeResult = runAgentWithModel(judgePrompt, judgeModel, logFile);
+          const judgeResult = await runAgentWithModel(judgePrompt, judgeModel, logFile);
 
           const verdict = parseJudgeVerdict(judgeResult.output);
           log(logFile, `[JUDGE] Verdict: ${verdict || "UNKNOWN"}`);

packages/server/src/utils/checkout-git.ts

@@ -1,5 +1,3 @@
-import { execFile } from "child_process";
-import { promisify } from "util";
 import { resolve, dirname, basename } from "path";
 import { existsSync, realpathSync } from "fs";
 import { open as openFile, stat as statFile } from "fs/promises";
@@ -8,10 +6,9 @@ import type { ParsedDiffFile } from "../server/utils/diff-highlighter.js";
 import { parseAndHighlightDiff } from "../server/utils/diff-highlighter.js";
 import { findExecutable } from "./executable.js";
 import { runGitCommand } from "./run-git-command.js";
+import { execCommand } from "./spawn.js";
 import { isPaseoOwnedWorktreeCwd } from "./worktree.js";
 import { requirePaseoWorktreeBaseRefName } from "./worktree-metadata.js";
-
-const execFileAsync = promisify(execFile);
 const READ_ONLY_GIT_ENV: NodeJS.ProcessEnv = {
   ...process.env,
   GIT_OPTIONAL_LOCKS: "0",
@@ -1899,7 +1896,7 @@ export async function createPullRequest(
   if (options.body) {
     args.push("-f", `body=${options.body}`);
   }
-  const { stdout } = await execFileAsync(ghPath, args, { cwd, env: ghEnv });
+  const { stdout } = await execCommand(ghPath, args, { cwd, env: ghEnv });
   const parsed = JSON.parse(stdout.trim());
   if (!parsed?.url || !parsed?.number) {
     throw new Error("GitHub CLI did not return PR url/number");
@@ -1951,7 +1948,7 @@ async function getPullRequestStatusUncached(cwd: string): Promise<PullRequestSta
     };
   }
   try {
-    const { stdout } = await execFileAsync(
+    const { stdout } = await execCommand(
       ghPath,
       ["pr", "view", "--json", "url,title,state,baseRefName,headRefName,mergedAt"],
       { cwd, env: { ...process.env, GIT_TERMINAL_PROMPT: "0" } },

packages/server/src/utils/executable.test.ts

@@ -1,78 +1,209 @@
-import { afterEach, beforeEach, describe, expect, test, vi } from "vitest";
+import { promisify } from "node:util";
+import { afterEach, describe, expect, test, vi } from "vitest";
 
-import { findExecutableSync, quoteWindowsArgument, quoteWindowsCommand } from "./executable.js";
+type ExecFileCallback = (error: Error | null, stdout: string, stderr: string) => void;
 
-type FindExecutableDependencies = NonNullable<Parameters<typeof findExecutableSync>[1]>;
+async function loadExecutableModule(params?: {
+  execFileImpl?: (
+    command: string,
+    args: string[],
+    options: unknown,
+    callback: ExecFileCallback,
+  ) => void;
+}) {
+  vi.resetModules();
 
-function createFindExecutableDependencies(): FindExecutableDependencies {
+  const execFileMock = vi.fn(
+    params?.execFileImpl ??
+      ((_command: string, _args: string[], _options: unknown, callback: ExecFileCallback) => {
+        callback(new Error("execFile not mocked"), "", "");
+      }),
+  );
+  Object.assign(execFileMock, {
+    [promisify.custom]: (command: string, args: string[], options: unknown) =>
+      new Promise<{ stdout: string; stderr: string }>((resolve, reject) => {
+        execFileMock(
+          command,
+          args,
+          options,
+          (error: Error | null, stdout: string, stderr: string) => {
+            if (error) {
+              reject(error);
+              return;
+            }
+            resolve({ stdout, stderr });
+          },
+        );
+      }),
+  });
+
+  vi.doMock("node:child_process", () => ({
+    execFile: execFileMock,
+  }));
+  const module = await import("./executable.js");
   return {
-    execFileSync: vi.fn(),
+    ...module,
-    existsSync: vi.fn(),
+    execFileMock,
-    platform: vi.fn(() => "darwin"),
   };
 }
 
-let findExecutableDependencies: FindExecutableDependencies;
+describe("findExecutable", () => {
+  const originalPlatform = process.platform;
 
-beforeEach(() => {
+  function setPlatform(value: string) {
-  findExecutableDependencies = createFindExecutableDependencies();
+    Object.defineProperty(process, "platform", { value, writable: true });
-});
+  }
 
-describe("findExecutableSync", () => {
+  afterEach(() => {
-  test("on Windows, resolves executables using where.exe with inherited PATH", () => {
+    setPlatform(originalPlatform);
-    findExecutableDependencies.platform = vi.fn(() => "win32");
+  });
-    findExecutableDependencies.execFileSync.mockReturnValue(
-      "C:\\Users\\boudr\\.local\\bin\\claude.exe\r\n",
-    );
 
-    expect(findExecutableSync("claude", findExecutableDependencies)).toBe(
+  test("on Windows, resolves executables using where.exe with inherited PATH", async () => {
+    setPlatform("win32");
+    const { execFileMock, findExecutable } = await loadExecutableModule({
+      execFileImpl: (_command, _args, _options, callback) => {
+        callback(null, "C:\\Users\\boudr\\.local\\bin\\claude.exe\r\n", "");
+      },
+    });
+
+    await expect(findExecutable("claude")).resolves.toBe(
       "C:\\Users\\boudr\\.local\\bin\\claude.exe",
     );
-    expect(findExecutableDependencies.execFileSync).toHaveBeenCalledOnce();
+    expect(execFileMock).toHaveBeenCalledOnce();
-    const call = findExecutableDependencies.execFileSync.mock.calls[0];
+    const call = execFileMock.mock.calls[0];
     expect(call?.[0]).toBe("where.exe");
     expect(call?.[1]).toEqual(["claude"]);
-    expect(call?.[2]?.encoding).toBe("utf8");
+    expect(call?.[2]).toMatchObject({
-    expect(call?.[2]?.windowsHide).toBe(true);
-  });
-
-  test("on Windows, preserves the first where.exe match", () => {
-    findExecutableDependencies.platform = vi.fn(() => "win32");
-    findExecutableDependencies.execFileSync.mockReturnValue(
-      "C:\\nvm4w\\nodejs\\codex\r\nC:\\nvm4w\\nodejs\\codex.cmd\r\n",
-    );
-
-    expect(findExecutableSync("codex", findExecutableDependencies)).toBe(
-      "C:\\nvm4w\\nodejs\\codex",
-    );
-  });
-
-  test("on Unix, uses the last line from which output", () => {
-    findExecutableDependencies.execFileSync.mockReturnValue("/usr/local/bin/codex\n");
-
-    expect(findExecutableSync("codex", findExecutableDependencies)).toBe("/usr/local/bin/codex");
-    expect(findExecutableDependencies.execFileSync).toHaveBeenCalledWith("which", ["codex"], {
       encoding: "utf8",
+      windowsHide: true,
     });
   });
 
-  test("warns and returns null when the final which line is not an absolute path", () => {
+  test("on Windows, prefers an executable match from where.exe output", async () => {
-    findExecutableDependencies.execFileSync.mockReturnValue("codex\n");
+    setPlatform("win32");
+    const { findExecutable } = await loadExecutableModule({
+      execFileImpl: (_command, _args, _options, callback) => {
+        callback(null, "C:\\nvm4w\\nodejs\\codex\r\nC:\\nvm4w\\nodejs\\codex.cmd\r\n", "");
+      },
+    });
+
+    await expect(findExecutable("codex")).resolves.toBe("C:\\nvm4w\\nodejs\\codex.cmd");
+  });
+
+  test("on Windows, prefers .exe over .cmd, .ps1, and extensionless candidates", async () => {
+    setPlatform("win32");
+    const { findExecutable } = await loadExecutableModule({
+      execFileImpl: (_command, _args, _options, callback) => {
+        callback(
+          null,
+          [
+            "C:\\nvm4w\\nodejs\\codex",
+            "C:\\nvm4w\\nodejs\\codex.ps1",
+            "C:\\nvm4w\\nodejs\\codex.cmd",
+            "C:\\nvm4w\\nodejs\\codex.exe",
+          ].join("\r\n"),
+          "",
+        );
+      },
+    });
+
+    await expect(findExecutable("codex")).resolves.toBe("C:\\nvm4w\\nodejs\\codex.exe");
+  });
+
+  test("on Windows, returns null when where.exe output is empty", async () => {
+    setPlatform("win32");
+    const { findExecutable } = await loadExecutableModule({
+      execFileImpl: (_command, _args, _options, callback) => {
+        callback(null, "\r\n", "");
+      },
+    });
+
+    await expect(findExecutable("codex")).resolves.toBeNull();
+  });
+
+  test("on Windows, falls back to the first extensionless candidate when needed", async () => {
+    setPlatform("win32");
+    const { findExecutable } = await loadExecutableModule({
+      execFileImpl: (_command, _args, _options, callback) => {
+        callback(null, "C:\\nvm4w\\nodejs\\codex\r\n", "");
+      },
+    });
+
+    await expect(findExecutable("codex")).resolves.toBe("C:\\nvm4w\\nodejs\\codex");
+  });
+
+  test("on Unix, uses the last line from which output", async () => {
+    const { execFileMock, findExecutable } = await loadExecutableModule({
+      execFileImpl: (_command, _args, _options, callback) => {
+        callback(null, "/usr/local/bin/codex\n", "");
+      },
+    });
+
+    await expect(findExecutable("codex")).resolves.toBe("/usr/local/bin/codex");
+    expect(execFileMock).toHaveBeenCalledWith(
+      "which",
+      ["codex"],
+      { encoding: "utf8" },
+      expect.any(Function),
+    );
+  });
+
+  test("warns and returns null when the final which line is not an absolute path", async () => {
     const warnSpy = vi.spyOn(console, "warn").mockImplementation(() => {});
+    const { findExecutable } = await loadExecutableModule({
+      execFileImpl: (_command, _args, _options, callback) => {
+        callback(null, "codex\n", "");
+      },
+    });
 
-    expect(findExecutableSync("codex", findExecutableDependencies)).toBeNull();
+    await expect(findExecutable("codex")).resolves.toBeNull();
     expect(warnSpy).toHaveBeenCalledOnce();
 
     warnSpy.mockRestore();
   });
 
-  test("returns direct paths when they exist", () => {
+  test("returns null when which lookup fails", async () => {
-    findExecutableDependencies.existsSync.mockReturnValue(true);
+    const { findExecutable } = await loadExecutableModule({
+      execFileImpl: (_command, _args, _options, callback) => {
+        callback(new Error("which failed"), "", "");
+      },
+    });
 
-    expect(findExecutableSync("/usr/local/bin/codex", findExecutableDependencies)).toBe(
+    await expect(findExecutable("codex")).resolves.toBeNull();
-      "/usr/local/bin/codex",
+  });
-    );
+});
-    expect(findExecutableDependencies.existsSync).toHaveBeenCalledWith("/usr/local/bin/codex");
+
+describe("executableExists", () => {
+  const originalPlatform = process.platform;
+
+  function setPlatform(value: string) {
+    Object.defineProperty(process, "platform", { value, writable: true });
+  }
+
+  afterEach(() => {
+    setPlatform(originalPlatform);
+  });
+
+  test("returns the path when it already exists", async () => {
+    const { executableExists } = await loadExecutableModule();
+    const exists = vi.fn((candidate: string) => candidate === "/usr/local/bin/codex");
+
+    expect(executableExists("/usr/local/bin/codex", exists)).toBe("/usr/local/bin/codex");
+  });
+
+  test("on Windows, falls back to .exe, .cmd, then .ps1 for extensionless paths", async () => {
+    setPlatform("win32");
+    const { executableExists } = await loadExecutableModule();
+    const exists = vi.fn((candidate: string) => candidate === "C:\\tools\\codex.cmd");
+
+    expect(executableExists("C:\\tools\\codex", exists)).toBe("C:\\tools\\codex.cmd");
+  });
+
+  test("returns null when no matching path exists", async () => {
+    const { executableExists } = await loadExecutableModule();
+    const exists = vi.fn(() => false);
+
+    expect(executableExists("/missing/codex", exists)).toBeNull();
   });
 });
 
@@ -87,27 +218,31 @@ describe("quoteWindowsCommand", () => {
     setPlatform(originalPlatform);
   });
 
-  test("quotes a Windows path with spaces", () => {
+  test("quotes a Windows path with spaces", async () => {
     setPlatform("win32");
+    const { quoteWindowsCommand } = await loadExecutableModule();
     expect(quoteWindowsCommand("C:\\Program Files\\Anthropic\\claude.exe")).toBe(
       '"C:\\Program Files\\Anthropic\\claude.exe"',
     );
   });
 
-  test("does not double-quote an already-quoted path", () => {
+  test("does not double-quote an already-quoted path", async () => {
     setPlatform("win32");
+    const { quoteWindowsCommand } = await loadExecutableModule();
     expect(quoteWindowsCommand('"C:\\Program Files\\Anthropic\\claude.exe"')).toBe(
       '"C:\\Program Files\\Anthropic\\claude.exe"',
     );
   });
 
-  test("returns the command unchanged when there are no spaces", () => {
+  test("returns the command unchanged when there are no spaces", async () => {
     setPlatform("win32");
+    const { quoteWindowsCommand } = await loadExecutableModule();
     expect(quoteWindowsCommand("C:\\nvm4w\\nodejs\\codex")).toBe("C:\\nvm4w\\nodejs\\codex");
   });
 
-  test("returns the command unchanged on non-Windows platforms", () => {
+  test("returns the command unchanged on non-Windows platforms", async () => {
     setPlatform("darwin");
+    const { quoteWindowsCommand } = await loadExecutableModule();
     expect(quoteWindowsCommand("/usr/local/bin/claude code")).toBe("/usr/local/bin/claude code");
   });
 });
@@ -123,27 +258,31 @@ describe("quoteWindowsArgument", () => {
     setPlatform(originalPlatform);
   });
 
-  test("quotes a Windows argument with spaces", () => {
+  test("quotes a Windows argument with spaces", async () => {
     setPlatform("win32");
+    const { quoteWindowsArgument } = await loadExecutableModule();
     expect(quoteWindowsArgument("C:\\Program Files\\Anthropic\\cli.js")).toBe(
       '"C:\\Program Files\\Anthropic\\cli.js"',
     );
   });
 
-  test("does not double-quote an already-quoted argument", () => {
+  test("does not double-quote an already-quoted argument", async () => {
     setPlatform("win32");
+    const { quoteWindowsArgument } = await loadExecutableModule();
     expect(quoteWindowsArgument('"C:\\Program Files\\Anthropic\\cli.js"')).toBe(
       '"C:\\Program Files\\Anthropic\\cli.js"',
     );
   });
 
-  test("returns the argument unchanged when there are no spaces", () => {
+  test("returns the argument unchanged when there are no spaces", async () => {
     setPlatform("win32");
+    const { quoteWindowsArgument } = await loadExecutableModule();
     expect(quoteWindowsArgument("--version")).toBe("--version");
   });
 
-  test("returns the argument unchanged on non-Windows platforms", () => {
+  test("returns the argument unchanged on non-Windows platforms", async () => {
     setPlatform("darwin");
+    const { quoteWindowsArgument } = await loadExecutableModule();
     expect(quoteWindowsArgument("/usr/local/bin/claude code")).toBe("/usr/local/bin/claude code");
   });
 });

packages/server/src/utils/executable.ts

@@ -1,15 +1,21 @@
-import { execFile, execFileSync } from "node:child_process";
+import { execFile } from "node:child_process";
 import { existsSync } from "node:fs";
-import { platform } from "node:os";
+import path, { extname } from "node:path";
-import path from "node:path";
 import { promisify } from "node:util";
 
 const execFileAsync = promisify(execFile);
 
-export interface FindExecutableDependencies {
+function pickBestWindowsCandidate(lines: string[]): string | null {
-  execFileSync: typeof execFileSync;
+  const candidates = lines.filter((line) => line.length > 0);
-  existsSync: typeof existsSync;
+  if (candidates.length === 0) return null;
-  platform: typeof platform;
+
+  const extPriority = [".exe", ".cmd", ".ps1"];
+  for (const ext of extPriority) {
+    const match = candidates.find((candidate) => candidate.toLowerCase().endsWith(ext));
+    if (match) return match;
+  }
+
+  return candidates[0] ?? null;
 }
 
 function resolveExecutableFromWhichOutput(
@@ -44,58 +50,18 @@ function resolveExecutableFromWhichOutput(
  * enriches it at startup via inheritLoginShellEnv(); on Windows, Electron
  * inherits the full user environment from Explorer.
  */
-export function findExecutableSync(
+export function executableExists(
-  name: string,
+  executablePath: string,
-  dependencies?: FindExecutableDependencies,
+  exists: typeof existsSync = existsSync,
 ): string | null {
-  const trimmed = name.trim();
+  if (exists(executablePath)) return executablePath;
-  if (!trimmed) {
+  if (process.platform === "win32" && !extname(executablePath)) {
+    for (const ext of [".exe", ".cmd", ".ps1"]) {
+      const candidate = executablePath + ext;
+      if (exists(candidate)) return candidate;
+    }
+  }
   return null;
-  }
-
-  const deps: FindExecutableDependencies = {
-    execFileSync,
-    existsSync,
-    platform,
-    ...dependencies,
-  };
-
-  if (trimmed.includes("/") || trimmed.includes("\\")) {
-    return deps.existsSync(trimmed) ? trimmed : null;
-  }
-
-  if (deps.platform() === "win32") {
-    try {
-      const out = deps
-        .execFileSync("where.exe", [trimmed], {
-          encoding: "utf8",
-          windowsHide: true,
-        })
-        .trim();
-      return (
-        out
-          .split(/\r?\n/)
-          .map((line) => line.trim())
-          .find((line) => line.length > 0) ?? null
-      );
-    } catch {
-      return null;
-    }
-  }
-
-  try {
-    return resolveExecutableFromWhichOutput(
-      trimmed,
-      deps.execFileSync("which", [trimmed], { encoding: "utf8" }).trim(),
-      "which",
-    );
-  } catch {
-    return null;
-  }
-}
-
-export function isCommandAvailableSync(command: string): boolean {
-  return findExecutableSync(command) !== null;
 }
 
 export async function findExecutable(name: string): Promise<string | null> {
@@ -105,21 +71,22 @@ export async function findExecutable(name: string): Promise<string | null> {
   }
 
   if (trimmed.includes("/") || trimmed.includes("\\")) {
-    return existsSync(trimmed) ? trimmed : null;
+    return executableExists(trimmed);
   }
 
-  if (platform() === "win32") {
+  if (process.platform === "win32") {
     try {
       const { stdout } = await execFileAsync("where.exe", [trimmed], {
         encoding: "utf8",
         windowsHide: true,
       });
       return (
+        pickBestWindowsCandidate(
           stdout
             .trim()
             .split(/\r?\n/)
-          .map((line) => line.trim())
+            .map((line) => line.trim()),
-          .find((line) => line.length > 0) ?? null
+        ) ?? null
       );
     } catch {
       return null;

packages/server/src/utils/spawn.test.ts

@@ -0,0 +1,54 @@
+import { mkdtempSync, realpathSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import path from "node:path";
+import { afterEach, describe, expect, test } from "vitest";
+
+import { execCommand } from "./spawn.js";
+
+describe("execCommand", () => {
+  const tempDirs: string[] = [];
+
+  afterEach(() => {
+    for (const tempDir of tempDirs) {
+      rmSync(tempDir, { recursive: true, force: true });
+    }
+    tempDirs.length = 0;
+  });
+
+  test("returns stdout and stderr for a successful command", async () => {
+    await expect(execCommand("echo", ["hello"])).resolves.toEqual({
+      stdout: "hello\n",
+      stderr: "",
+    });
+  });
+
+  test("rejects when the command times out", async () => {
+    const command =
+      process.platform === "win32"
+        ? {
+            command: process.execPath,
+            args: ["-e", "setTimeout(() => {}, 10_000)"],
+          }
+        : { command: "sleep", args: ["10"] };
+
+    await expect(execCommand(command.command, command.args, { timeout: 100 })).rejects.toThrow();
+  });
+
+  test("runs the command in the provided cwd", async () => {
+    const cwd = realpathSync(mkdtempSync(path.join(tmpdir(), "spawn-test-")));
+    tempDirs.push(cwd);
+
+    const command =
+      process.platform === "win32"
+        ? {
+            command: process.execPath,
+            args: ["-e", "console.log(process.cwd())"],
+          }
+        : { command: "pwd", args: [] };
+
+    await expect(execCommand(command.command, command.args, { cwd })).resolves.toEqual({
+      stdout: `${cwd}\n`,
+      stderr: "",
+    });
+  });
+});

packages/server/src/utils/spawn.ts

@@ -1,15 +1,27 @@
-import { spawn, type ChildProcess, type SpawnOptions } from "node:child_process";
+import { execFile, spawn, type ChildProcess, type SpawnOptions } from "node:child_process";
+import { promisify } from "node:util";
+
+const execFileAsync = promisify(execFile);
+
+interface ExecCommandOptions {
+  cwd?: string;
+  env?: NodeJS.ProcessEnv;
+  encoding?: BufferEncoding;
+  timeout?: number;
+  maxBuffer?: number;
+}
+
+interface ExecCommandResult {
+  stdout: string;
+  stderr: string;
+}
+
+function quoteForCmd(value: string): string {
+  if (!value.includes(" ")) return value;
+  if (value.startsWith('"') && value.endsWith('"')) return value;
+  return `"${value}"`;
+}
 
-/**
- * Platform-aware spawn that centralizes Windows shell and quoting concerns.
- *
- * On Windows:
- * - Enables `shell: true` (routes through cmd.exe) unless the caller explicitly sets `shell`
- * - Quotes the command and arguments so paths with spaces survive cmd.exe parsing
- * - Always sets `windowsHide: true` to prevent console window flashes
- *
- * On other platforms the call is passed through to `spawn` unchanged (with `windowsHide: true`).
- */
 export function spawnProcess(
   command: string,
   args: string[],
@@ -27,12 +39,38 @@ export function spawnProcess(
   });
 }
 
-/**
+export async function execCommand(
- * Quote a string for cmd.exe if it contains spaces and isn't already quoted.
+  command: string,
- * No-op for strings without spaces or strings that are already double-quoted.
+  args: string[],
- */
+  options?: ExecCommandOptions,
-function quoteForCmd(value: string): string {
+): Promise<ExecCommandResult> {
-  if (!value.includes(" ")) return value;
+  const isWindows = process.platform === "win32";
-  if (value.startsWith('"') && value.endsWith('"')) return value;
+  const resolvedCommand = isWindows ? quoteForCmd(command) : command;
-  return `"${value}"`;
+  const resolvedArgs = isWindows ? args.map(quoteForCmd) : args;
+
+  return execFileAsync(resolvedCommand, resolvedArgs, {
+    cwd: options?.cwd,
+    env: options?.env,
+    encoding: options?.encoding ?? "utf8",
+    timeout: options?.timeout,
+    maxBuffer: options?.maxBuffer,
+    shell: isWindows,
+    windowsHide: true,
+  }) as Promise<ExecCommandResult>;
+}
+
+export function platformShell(): { command: string; flag: string[] } {
+  if (process.platform === "win32") {
+    return { command: "cmd.exe", flag: ["/c"] };
+  }
+
+  return { command: "/bin/sh", flag: ["-lc"] };
+}
+
+export function platformBash(): { command: string; flag: string[] } {
+  if (process.platform === "win32") {
+    return { command: "cmd.exe", flag: ["/c"] };
+  }
+
+  return { command: "/bin/bash", flag: ["-lc"] };
 }

packages/server/src/utils/worktree.ts

@@ -1,4 +1,4 @@
-import { exec, spawn } from "child_process";
+import { exec } from "child_process";
 import { promisify } from "util";
 import { existsSync, mkdirSync, readFileSync, realpathSync, rmSync, statSync } from "fs";
 import { join, basename, dirname, resolve, sep } from "path";
@@ -13,6 +13,7 @@ import {
   writePaseoWorktreeRuntimeMetadata,
 } from "./worktree-metadata.js";
 import { runGitCommand } from "./run-git-command.js";
+import { platformBash, spawnProcess } from "./spawn.js";
 import { resolvePaseoHome } from "../server/paseo-home.js";
 
 interface PaseoConfig {
@@ -204,7 +205,7 @@ async function execSetupCommand(
     const { stdout, stderr } = await execAsync(command, {
       cwd: options.cwd,
       env: options.env,
-      shell: "/bin/bash",
+      ...(process.platform === "win32" ? {} : { shell: "/bin/bash" }),
     });
     return {
       command,
@@ -275,7 +276,8 @@ async function execSetupCommandStreamed(options: {
       cwd: options.cwd,
     });
 
-    const child = spawn("/bin/bash", ["-lc", options.command], {
+    const shell = platformBash();
+    const child = spawnProcess(shell.command, [...shell.flag, options.command], {
       cwd: options.cwd,
       env: options.env,
       stdio: ["ignore", "pipe", "pipe"],