@pushpalsdev/cli 1.0.23 → 1.0.24

package/dist/pushpals-cli.js

@@ -2158,6 +2158,52 @@ function stopRuntimeServices(services) {
     } catch {}
   }
 }
+function resolveGracefulShutdownPriority(name) {
+  if (name === "source_control_manager")
+    return 0;
+  if (name === "remotebuddy")
+    return 1;
+  if (name === "localbuddy")
+    return 2;
+  return 3;
+}
+async function waitForRuntimeServicesExit(services, timeoutMs) {
+  if (services.length === 0)
+    return true;
+  const deadline = Date.now() + Math.max(0, timeoutMs);
+  while (Date.now() < deadline) {
+    if (services.every((service) => service.exited))
+      return true;
+    await Bun.sleep(100);
+  }
+  return services.every((service) => service.exited);
+}
+async function stopRuntimeServicesGracefully(services, timeoutMs = 1e4) {
+  if (services.length === 0)
+    return;
+  const running = services.filter((service) => !service.exited);
+  if (running.length === 0)
+    return;
+  const ordered = [...running].sort((a, b) => resolveGracefulShutdownPriority(a.name) - resolveGracefulShutdownPriority(b.name));
+  const nonServer = ordered.filter((service) => service.name !== "server");
+  const server = ordered.filter((service) => service.name === "server");
+  for (const service of nonServer) {
+    try {
+      service.proc.kill("SIGTERM");
+    } catch {}
+  }
+  await waitForRuntimeServicesExit(nonServer, Math.max(1000, timeoutMs - 2000));
+  for (const service of server) {
+    try {
+      service.proc.kill("SIGTERM");
+    } catch {}
+  }
+  await waitForRuntimeServicesExit(server, Math.min(3000, timeoutMs));
+  const remaining = ordered.filter((service) => !service.exited);
+  if (remaining.length > 0) {
+    stopRuntimeServices(remaining);
+  }
+}
 function prependExecutableDirToPath(env, executablePath, platform = process.platform) {
   const resolvedPath = String(executablePath ?? "").trim();
   if (!resolvedPath)
@@ -2704,7 +2750,7 @@ function removeCliClearTarget(target) {
     };
   }
 }
-async function requestLocalRuntimeShutdownForClear(serverUrl, repoRoot) {
+async function requestLocalRuntimeShutdown(serverUrl, repoRoot, reason) {
   if (!await probeServer(serverUrl)) {
     return { attempted: false, accepted: false };
   }
@@ -2721,7 +2767,7 @@ async function requestLocalRuntimeShutdownForClear(serverUrl, repoRoot) {
     const response = await fetchWithTimeout(`${serverUrl}/admin/shutdown`, {
       method: "POST",
       headers: { "Content-Type": "application/json" },
-      body: JSON.stringify({ reason: "pushpals --clear" })
+      body: JSON.stringify({ reason })
     }, 5000);
     if (!response.ok) {
       const detail = await response.text().catch(() => "");
@@ -2742,7 +2788,7 @@ async function requestLocalRuntimeShutdownForClear(serverUrl, repoRoot) {
 }
 async function clearPushpalsState(opts) {
   console.log("[pushpals] Clear requested. Removing repo-local PushPals state.");
-  const shutdown = await requestLocalRuntimeShutdownForClear(opts.serverUrl, opts.repoRoot);
+  const shutdown = await requestLocalRuntimeShutdown(opts.serverUrl, opts.repoRoot, "pushpals --clear");
   if (shutdown.attempted && shutdown.accepted) {
     console.log("[pushpals] Local runtime shutdown accepted; waiting for services to exit...");
     await Bun.sleep(1500);
@@ -3879,6 +3925,22 @@ async function main() {
     stopRuntimeServices(autoStartedServices);
     autoStartedServices = [];
   };
+  const stopAutoStartedServicesGracefully = async (reason) => {
+    if (autoStartedServices.length === 0)
+      return;
+    const services = autoStartedServices;
+    autoStartedServices = [];
+    const shutdown = await requestLocalRuntimeShutdown(serverUrl, repoRoot, reason);
+    if (shutdown.attempted && shutdown.accepted) {
+      console.log("[pushpals] Local runtime shutdown accepted; waiting for services to exit...");
+      await Bun.sleep(1500);
+    } else if (shutdown.attempted) {
+      console.warn(`[pushpals] Local runtime shutdown request was not accepted${shutdown.detail ? `: ${shutdown.detail}` : "."}`);
+    } else if (shutdown.detail) {
+      console.warn(`[pushpals] ${shutdown.detail}`);
+    }
+    await stopRuntimeServicesGracefully(services);
+  };
   let serverHealthy = await probeServer(serverUrl);
   const serverWasAlreadyHealthy = serverHealthy;
   if (!serverHealthy && workerpalDockerPrecheck.status === "failed") {
@@ -4046,26 +4108,36 @@ ${line}
     console.log(line);
   };
   const streamTask = parsed.noStream ? Promise.resolve() : parsed.runtimeOnly ? Promise.resolve() : runSessionStream(serverUrl, activeSessionId, cliClient, printIncoming, streamAbort.signal);
-  let shuttingDown = false;
+  let stopPromise = null;
   const requestStop = () => {
-    if (shuttingDown)
-      return;
-    shuttingDown = true;
-    console.log("[pushpals] Shutting down CLI session...");
-    streamAbort.abort();
-    if (rl)
-      rl.close();
-    try {
-      monitoringHub?.stop();
-    } catch {}
-    if (autoStartedServices.length > 0) {
-      console.log("[pushpals] Stopping embedded runtime services...");
-    }
-    stopAutoStartedServices();
+    if (stopPromise)
+      return stopPromise;
+    stopPromise = (async () => {
+      console.log("[pushpals] Shutting down CLI session...");
+      streamAbort.abort();
+      const activeRl = rl;
+      rl = null;
+      if (activeRl)
+        activeRl.close();
+      try {
+        monitoringHub?.stop();
+      } catch {}
+      if (autoStartedServices.length > 0) {
+        console.log("[pushpals] Stopping embedded runtime services...");
+      }
+      await stopAutoStartedServicesGracefully("pushpals CLI exit");
+    })();
+    return stopPromise;
   };
-  process.once("SIGINT", requestStop);
-  process.once("SIGTERM", requestStop);
-  process.once("exit", requestStop);
+  process.once("SIGINT", () => {
+    requestStop();
+  });
+  process.once("SIGTERM", () => {
+    requestStop();
+  });
+  process.once("exit", () => {
+    stopAutoStartedServices();
+  });
   if (parsed.runtimeOnly) {
     console.log("[pushpals] Runtime-only mode is active. Send `exit` on stdin or terminate the process to stop.");
     await new Promise((resolveStop) => {
@@ -4095,7 +4167,7 @@ ${line}
         finish();
       });
     });
-    requestStop();
+    await requestStop();
     await Promise.race([streamTask, Bun.sleep(2000)]);
     return;
   }
@@ -4113,7 +4185,7 @@ ${line}
       continue;
     }
     if (isCliExitCommand(text)) {
-      requestStop();
+      await requestStop();
       break;
     }
     if (text === "/hub") {
@@ -4153,7 +4225,7 @@ ${line}
     }
     rl.prompt();
   }
-  requestStop();
+  await requestStop();
   await Promise.race([streamTask, Bun.sleep(2000)]);
 }
 if (import.meta.main) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pushpalsdev/cli",
-  "version": "1.0.23",
+  "version": "1.0.24",
   "description": "PushPals terminal CLI for LocalBuddy -> RemoteBuddy orchestration",
   "license": "MIT",
   "repository": {

package/runtime/configs/default.toml

@@ -198,7 +198,7 @@ session_id = "workerpals-dev"
 [workerpals.openai_codex]
 timeout_ms = 7200000
 progress_log_interval_s = 30
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 approval_policy = "never"
 sandbox = "workspace-write"
 color = "never"

package/runtime/configs/local.example.toml

@@ -12,7 +12,7 @@ model = "gpt-5.4"
 codex_auth_mode = "chatgpt"
 codex_bin = "bun x --yes @openai/codex"
 codex_timeout_ms = 120000
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 
 [remotebuddy.llm]
 backend = "openai_codex"
@@ -20,7 +20,7 @@ model = "gpt-5.4"
 codex_auth_mode = "chatgpt"
 codex_bin = "bun x --yes @openai/codex"
 codex_timeout_ms = 120000
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 
 [remotebuddy]
 max_workerpals = 10
@@ -46,7 +46,7 @@ model = "gpt-5.4"
 codex_auth_mode = "chatgpt"
 codex_bin = "bun x --yes @openai/codex"
 codex_timeout_ms = 120000
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 
 [workerpals]
 executor = "openai_codex"
@@ -92,7 +92,7 @@ bin = "bun x --yes @openai/codex"
 timeout_ms = 7200000
 progress_log_interval_s = 30
 # timeout_s = 120  # optional; if set, overrides timeout_ms
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 approval_policy = "never"
 sandbox = "workspace-write"
 color = "never"

package/runtime/sandbox/apps/workerpals/src/backends/openai_codex/openai_codex_executor.py

@@ -83,6 +83,18 @@ _VALID_AUTH_MODES = {"auto", "api_key", "chatgpt"}
 _VALID_REASONING_EFFORTS = {"low", "medium", "high", "xhigh"}
 
 
+def _model_supports_xhigh_reasoning(model: str) -> bool:
+    normalized = str(model or "").strip().lower()
+    if not normalized:
+        return False
+    return not (
+        normalized == "gpt-5.4"
+        or normalized.startswith("gpt-5.4-")
+        or normalized == "codex-1p"
+        or normalized.startswith("codex-1p-")
+    )
+
+
 @dataclass(frozen=True)
 class OpenAICodexRuntimeConfig:
     codex_bin_json: str
@@ -152,7 +164,7 @@ class OpenAICodexRuntimeConfig:
             reasoning_effort=cfg.get_str(
                 env_names=("WORKERPALS_LLM_REASONING_EFFORT", "WORKERPALS_OPENAI_CODEX_REASONING_EFFORT"),
                 config_paths=("workerpals.llm.reasoning_effort", "workerpals.openai_codex.reasoning_effort"),
-                default="xhigh",
+                default="high",
             ),
             approval_policy=cfg.get_str(
                 env_names=("WORKERPALS_OPENAI_CODEX_APPROVAL_POLICY",),
@@ -316,18 +328,23 @@ def _resolve_communicate_timeout_seconds(config: OpenAICodexRuntimeConfig) -> Op
     return max(1, timeout_ms // 1000)
 
 
-def _resolve_reasoning_effort(config: OpenAICodexRuntimeConfig) -> str:
+def _resolve_reasoning_effort(config: OpenAICodexRuntimeConfig, model: str = DEFAULT_CODEX_MODEL) -> str:
     raw = config.reasoning_effort
     normalized = str(raw).strip().lower()
     if normalized in {"extra high", "extra-high", "extrahigh", "x-high"}:
         normalized = "xhigh"
+    if normalized == "xhigh" and not _model_supports_xhigh_reasoning(model):
+        log.info(
+            f"Downgrading workerpals.openai_codex.reasoning_effort='xhigh' to 'high' for model {model!r}."
+        )
+        return "high"
     if normalized in _VALID_REASONING_EFFORTS:
         return normalized
     log.info(
         "Invalid workerpals.openai_codex.reasoning_effort="
-        f"{raw!r}; using default 'xhigh'. Allowed: low, medium, high, xhigh."
+        f"{raw!r}; using default 'high'. Allowed: low, medium, high, xhigh."
     )
-    return "xhigh"
+    return "high"
 
 
 def _resolve_progress_log_interval_seconds(config: OpenAICodexRuntimeConfig) -> int:
@@ -1006,7 +1023,7 @@ def _run_codex_task(
     )
     # JSON event output is noisy by default; prefer plain text + output-last-message.
     use_json = runtime_config.json_output
-    reasoning_effort = _resolve_reasoning_effort(runtime_config)
+    reasoning_effort = _resolve_reasoning_effort(runtime_config, model)
     communicate_timeout_s = _resolve_communicate_timeout_seconds(runtime_config)
     prompt = _build_instruction(instruction, supplemental_guidance)
     baseline_changes = summarize_git_changes(repo)

package/runtime/sandbox/apps/workerpals/src/backends/openai_codex/test_openai_codex_runtime_config.py

@@ -60,17 +60,26 @@ class OpenAICodexRuntimeConfigTests(unittest.TestCase):
         self.assertEqual(cfg.approval_policy, "never")
         self.assertEqual(cfg.sandbox, "workspace-write")
         self.assertEqual(cfg.color, "never")
-        self.assertEqual(cfg.reasoning_effort, "xhigh")
+        self.assertEqual(cfg.reasoning_effort, "high")
         self.assertFalse(cfg.json_output)
 
-    def test_reasoning_effort_accepts_extra_high_alias(self) -> None:
+    def test_reasoning_effort_caps_extra_high_for_gpt_5_4(self) -> None:
         cfg = OpenAICodexRuntimeConfig.from_sources(
             SettingsResolver(
                 env={"WORKERPALS_OPENAI_CODEX_REASONING_EFFORT": "extra high"},
                 config_loader=lambda: {},
             ),
         )
-        self.assertEqual(_resolve_reasoning_effort(cfg), "xhigh")
+        self.assertEqual(_resolve_reasoning_effort(cfg), "high")
+
+    def test_reasoning_effort_preserves_extra_high_for_future_models(self) -> None:
+        cfg = OpenAICodexRuntimeConfig.from_sources(
+            SettingsResolver(
+                env={"WORKERPALS_OPENAI_CODEX_REASONING_EFFORT": "extra high"},
+                config_loader=lambda: {},
+            ),
+        )
+        self.assertEqual(_resolve_reasoning_effort(cfg, model="gpt-6-preview"), "xhigh")
 
     def test_runtime_config_prefers_explicit_config_dir_override(self) -> None:
         import executor_base

package/runtime/sandbox/apps/workerpals/src/docker_executor.ts

@@ -1048,6 +1048,7 @@ export class DockerExecutor {
 
     const worktreeRelPath = relative(this.options.repo, worktreePath).replace(/\\/g, "/");
     const containerWorktreePath = `/repo/${worktreeRelPath}`;
+    await this.waitForWorktreePathInWarmContainer(containerWorktreePath);
 
     const args: string[] = [
       "exec",
@@ -1124,6 +1125,26 @@ export class DockerExecutor {
     return result;
   }
 
+  private async waitForWorktreePathInWarmContainer(
+    containerWorktreePath: string,
+    timeoutMs = 5_000,
+  ): Promise<void> {
+    const deadline = Date.now() + timeoutMs;
+    let lastDetail = "";
+    const command = `test -d ${shellSingleQuote(containerWorktreePath)}`;
+    while (Date.now() < deadline) {
+      const result = await this.runWarmShell(command);
+      if (result.ok) return;
+      lastDetail = [result.stdout, result.stderr].filter(Boolean).join("\n").trim();
+      await this.sleep(100);
+    }
+    throw new Error(
+      `worktree path not visible inside warm container after ${timeoutMs}ms: ${containerWorktreePath}${
+        lastDetail ? ` (${lastDetail})` : ""
+      }`,
+    );
+  }
+
   private normalizeProvider(raw: string): string {
     const value = raw.trim().toLowerCase();
     if (!value) return "auto";
@@ -1450,6 +1471,9 @@ export class DockerExecutor {
       /\btemporary failure\b/i,
       /\bopenhands wrapper timed out\b/i,
       /\bjob timed out in docker executor\b/i,
+      /\bworktree path not visible inside warm container\b/i,
+      /\bchdir to cwd\b/i,
+      /\bunable to start container process\b/i,
     ];
     return transientPatterns.some((pattern) => pattern.test(text));
   }

package/runtime/sandbox/apps/workerpals/src/execute_job.ts

@@ -2024,17 +2024,19 @@ export function shouldUseCodexCliForExecutor(executor: string): boolean {
 
 function normalizeCodexReasoningEffort(
   value: unknown,
+  model = "",
 ): "low" | "medium" | "high" | "xhigh" {
   const normalized = String(value ?? "")
     .trim()
     .toLowerCase();
+  const supportsExtraHigh = !/^(gpt-5\.4(?:$|-)|codex-1p(?:$|-))/i.test(String(model ?? "").trim());
   if (
     normalized === "low" ||
     normalized === "medium" ||
     normalized === "high" ||
     normalized === "xhigh"
   ) {
-    return normalized;
+    return normalized === "xhigh" && !supportsExtraHigh ? "high" : normalized;
   }
   if (
     normalized === "extra high" ||
@@ -2042,9 +2044,9 @@ function normalizeCodexReasoningEffort(
     normalized === "extrahigh" ||
     normalized === "x-high"
   ) {
-    return "xhigh";
+    return supportsExtraHigh ? "xhigh" : "high";
   }
-  return "xhigh";
+  return "high";
 }
 
 async function generateCommitMessageFromDiff(
@@ -2105,6 +2107,7 @@ async function generateCommitMessageFromDiffViaCodex(
   })();
   const reasoningEffort = normalizeCodexReasoningEffort(
     runtimeConfig.workerpals.llm.reasoningEffort,
+    model,
   );
   const tmpOutputPath = resolve(
     Bun.env.TEMP || Bun.env.TMP || Bun.env.TMPDIR || "/tmp",

package/runtime/sandbox/configs/default.toml

@@ -198,7 +198,7 @@ session_id = "workerpals-dev"
 [workerpals.openai_codex]
 timeout_ms = 7200000
 progress_log_interval_s = 30
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 approval_policy = "never"
 sandbox = "workspace-write"
 color = "never"

package/runtime/sandbox/configs/local.example.toml

@@ -12,7 +12,7 @@ model = "gpt-5.4"
 codex_auth_mode = "chatgpt"
 codex_bin = "bun x --yes @openai/codex"
 codex_timeout_ms = 120000
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 
 [remotebuddy.llm]
 backend = "openai_codex"
@@ -20,7 +20,7 @@ model = "gpt-5.4"
 codex_auth_mode = "chatgpt"
 codex_bin = "bun x --yes @openai/codex"
 codex_timeout_ms = 120000
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 
 [remotebuddy]
 max_workerpals = 10
@@ -46,7 +46,7 @@ model = "gpt-5.4"
 codex_auth_mode = "chatgpt"
 codex_bin = "bun x --yes @openai/codex"
 codex_timeout_ms = 120000
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 
 [workerpals]
 executor = "openai_codex"
@@ -92,7 +92,7 @@ bin = "bun x --yes @openai/codex"
 timeout_ms = 7200000
 progress_log_interval_s = 30
 # timeout_s = 120  # optional; if set, overrides timeout_ms
-reasoning_effort = "xhigh"
+reasoning_effort = "high"
 approval_policy = "never"
 sandbox = "workspace-write"
 color = "never"