@pushpalsdev/cli 1.1.35 → 1.1.37

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pushpalsdev/cli",
-  "version": "1.1.35",
+  "version": "1.1.37",
   "description": "PushPals terminal CLI for LocalBuddy -> RemoteBuddy orchestration",
   "license": "MIT",
   "repository": {

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

@@ -116,6 +116,7 @@ _WEB_REVIEW_NO_EDIT_WATCHDOG_S = 240
 _BACKGROUND_NO_EDIT_WATCHDOG_S = 120
 _NO_EDIT_RECOVERY_WATCHDOG_S = 90
 _DEFAULT_NO_EDIT_RECHECK_S = 120
+_DEFAULT_NO_EDIT_COMMAND_GRACE_S = 240
 _DEFAULT_STARTUP_STALL_WATCHDOG_S = 210
 _RECOVERY_STARTUP_STALL_WATCHDOG_S = 150
 _DEFAULT_ROLLOUT_WATCHDOG_S = 300
@@ -757,6 +758,27 @@ def _resolve_no_edit_recheck_seconds(communicate_timeout_s: Optional[int]) -> in
     return max(1, min(_DEFAULT_NO_EDIT_RECHECK_S, upper))
 
 
+def _resolve_no_edit_command_grace_seconds(communicate_timeout_s: Optional[int]) -> Optional[int]:
+    if not communicate_timeout_s:
+        return None
+
+    raw = os.environ.get("WORKERPALS_OPENAI_CODEX_NO_EDIT_COMMAND_GRACE_S", "").strip()
+    if raw:
+        if raw == "0":
+            return None
+        parsed = _to_positive_int(raw)
+        if parsed is None:
+            log.info(
+                "Invalid WORKERPALS_OPENAI_CODEX_NO_EDIT_COMMAND_GRACE_S="
+                f"{raw!r}; using default command-progress grace."
+            )
+        else:
+            return max(1, min(parsed, max(1, communicate_timeout_s - 1)))
+
+    upper = max(1, communicate_timeout_s - 1)
+    return max(1, min(_DEFAULT_NO_EDIT_COMMAND_GRACE_S, upper))
+
+
 def _resolve_startup_stall_watchdog_seconds(
     communicate_timeout_s: Optional[int],
     recovery_attempt: int = 0,
@@ -1339,10 +1361,94 @@ def _empty_codex_trace() -> Dict[str, Any]:
         "prompt_tokens": 0,
         "completion_tokens": 0,
         "total_tokens": 0,
+        "active_command_ids": [],
+        "command_event_count": 0,
+        "last_command_activity_at": None,
+        "last_command_summary": "",
     }
 
 
-def _record_live_codex_stdout_line(line: str, use_json: bool, trace: Dict[str, Any]) -> None:
+def _looks_like_codex_command_item(value: Any) -> bool:
+    if not isinstance(value, dict):
+        return False
+    type_text = " ".join(
+        str(value.get(key) or "")
+        for key in ("type", "item_type", "kind", "name", "tool_name")
+    ).lower()
+    if any(marker in type_text for marker in ("command_execution", "exec_command", "shell_command")):
+        return True
+    return any(key in value for key in ("command", "cmd", "exit_code", "aggregated_output"))
+
+
+def _record_codex_command_activity(
+    parsed: Dict[str, Any],
+    event_type: str,
+    trace: Dict[str, Any],
+    now: float,
+) -> None:
+    item = parsed.get("item")
+    command_source: Any = item if _looks_like_codex_command_item(item) else parsed
+    if not _looks_like_codex_command_item(command_source):
+        return
+
+    command_text = ""
+    if isinstance(command_source, dict):
+        for key in ("command", "cmd", "name"):
+            raw = command_source.get(key)
+            if isinstance(raw, str) and raw.strip():
+                command_text = _truncate_inline(raw.strip(), 160)
+                break
+    command_id = ""
+    if isinstance(command_source, dict):
+        command_id = str(
+            command_source.get("id")
+            or command_source.get("call_id")
+            or command_source.get("item_id")
+            or command_text
+            or "command"
+        ).strip()
+    command_id = command_id or "command"
+
+    active = trace.setdefault("active_command_ids", [])
+    if not isinstance(active, list):
+        active = []
+        trace["active_command_ids"] = active
+
+    status_text = ""
+    if isinstance(command_source, dict):
+        status_text = " ".join(
+            str(command_source.get(key) or "")
+            for key in ("status", "state", "outcome")
+        ).lower()
+    event_lower = event_type.lower()
+    completed = (
+        "completed" in event_lower
+        or "failed" in event_lower
+        or "error" in event_lower
+        or any(marker in status_text for marker in ("completed", "failed", "cancelled", "canceled", "exited"))
+    )
+    started = (
+        "started" in event_lower
+        or "updated" in event_lower
+        or any(marker in status_text for marker in ("running", "in_progress", "started"))
+    )
+
+    if completed:
+        trace["active_command_ids"] = [item for item in active if str(item) != command_id]
+    elif started and command_id not in active:
+        active.append(command_id)
+
+    trace["command_event_count"] = to_int(trace.get("command_event_count"), 0) + 1
+    trace["last_command_activity_at"] = float(now)
+    trace["last_command_summary"] = command_text or event_type
+
+
+def _record_live_codex_stdout_line(
+    line: str,
+    use_json: bool,
+    trace: Dict[str, Any],
+    now: Optional[float] = None,
+) -> None:
     stripped = line.strip()
     if not stripped:
         return
@@ -1369,6 +1475,7 @@ def _record_live_codex_stdout_line(line: str, use_json: bool, trace: Dict[str, A
             return
 
         if isinstance(parsed, dict):
+            observed_at = float(now if now is not None else time.monotonic())
             usage = _extract_usage_counts(parsed)
             if usage is not None:
                 trace["prompt_tokens"] = max(
@@ -1385,6 +1492,7 @@ def _record_live_codex_stdout_line(line: str, use_json: bool, trace: Dict[str, A
                 .strip()
                 or "event"
             )
+            _record_codex_command_activity(parsed, event_type, trace, observed_at)
             event_type_counts[event_type] = to_int(event_type_counts.get(event_type), 0) + 1
             summary = _summarize_json_event(parsed)
             # Reasoning can arrive under generic event types (for example item.updated).
@@ -1449,10 +1557,13 @@ def _finalize_codex_stdout_trace(trace: Dict[str, Any], use_json: bool) -> Dict[
     prompt_tokens = to_int(trace.get("prompt_tokens"), 0)
     completion_tokens = to_int(trace.get("completion_tokens"), 0)
     total_tokens = to_int(trace.get("total_tokens"), 0)
+    command_event_count = to_int(trace.get("command_event_count"), 0)
     if reasoning_events > 0:
         log.info(f"[codex] Reasoning-like event(s): {reasoning_events}")
     elif use_json and valid_json > 0:
         log.info("[codex] No reasoning-like events observed in this run.")
+    if command_event_count > 0:
+        log.info(f"[codex] Command execution event(s): {command_event_count}")
     if total_tokens > 0:
         log.info(
             f"[codex] Usage observed: prompt={prompt_tokens} completion={completion_tokens} total={total_tokens}"
@@ -1473,6 +1584,7 @@ def _finalize_codex_stdout_trace(trace: Dict[str, Any], use_json: bool) -> Dict[
         "prompt_tokens": prompt_tokens,
         "completion_tokens": completion_tokens,
         "total_tokens": total_tokens,
+        "command_event_count": command_event_count,
     }
 
 
@@ -2299,8 +2411,14 @@ def _run_codex_task(
                         if not line:
                             continue
                         with trace_lock:
-                            last_activity_at["ts"] = time.monotonic()
-                            _record_live_codex_stdout_line(line, use_json, stdout_trace_state)
+                            observed_at = time.monotonic()
+                            last_activity_at["ts"] = observed_at
+                            _record_live_codex_stdout_line(
+                                line,
+                                use_json,
+                                stdout_trace_state,
+                                observed_at,
+                            )
                 except Exception:
                     pass
                 finally:
@@ -2377,6 +2495,7 @@ def _run_codex_task(
                 else None
             )
             no_edit_recheck_s = _resolve_no_edit_recheck_seconds(communicate_timeout_s)
+            no_edit_command_grace_s = _resolve_no_edit_command_grace_seconds(communicate_timeout_s)
             startup_stall_watchdog_s = _resolve_startup_stall_watchdog_seconds(
                 communicate_timeout_s,
                 recovery_attempt=startup_stall_recovery_attempt,
@@ -2405,6 +2524,9 @@ def _run_codex_task(
                 if rollout_watchdog_s is not None
                 else None
             )
+            publishable_progress_seen_at: Optional[float] = None
+            publishable_progress_finalized = False
+            publishable_progress_paths: List[str] = []
 
             while proc.poll() is None:
                 now = time.monotonic()
@@ -2457,6 +2579,44 @@ def _run_codex_task(
                                 "before startup-stall recovery."
                             )
                             continue
+                        command_event_count = to_int(live_trace.get("command_event_count"), 0)
+                        active_commands_raw = live_trace.get("active_command_ids")
+                        active_command_count = (
+                            len(active_commands_raw)
+                            if isinstance(active_commands_raw, list)
+                            else 0
+                        )
+                        last_command_activity_at = 0.0
+                        try:
+                            last_command_activity_at = float(
+                                live_trace.get("last_command_activity_at") or 0.0
+                            )
+                        except Exception:
+                            last_command_activity_at = 0.0
+                        if command_event_count > 0 and no_edit_command_grace_s is not None:
+                            command_grace_deadline = 0.0
+                            if active_command_count > 0:
+                                # Do not kill while Codex is actively running a tool command; poll
+                                # again soon, but keep the total grace bounded by the hard cap below.
+                                command_grace_deadline = now + min(60.0, float(no_edit_command_grace_s))
+                            elif last_command_activity_at > 0:
+                                command_grace_deadline = last_command_activity_at + float(
+                                    no_edit_command_grace_s
+                                )
+                            if command_grace_deadline > now:
+                                no_edit_deadline = command_grace_deadline
+                                remaining_s = int(max(1.0, command_grace_deadline - now))
+                                command_detail = (
+                                    f"{active_command_count} active command(s)"
+                                    if active_command_count > 0
+                                    else "recent command completion"
+                                )
+                                log.info(
+                                    "No-edit watchdog observed Codex tool progress "
+                                    f"({command_detail}); allowing {remaining_s}s for a "
+                                    "publishable patch before recovery."
+                                )
+                                continue
                         no_edit_artifact_only_paths = _describe_non_publishable_paths(
                             changed_paths,
                             baseline_snapshot,
@@ -2478,6 +2638,22 @@ def _run_codex_task(
                             )
                         _terminate_active_child()
                         break
+                    if publishable_progress_seen_at is None:
+                        publishable_progress_seen_at = now
+                        publishable_progress_paths = list(effective_paths)
+                    elif _has_credible_shell_wrapper_progress(effective_paths):
+                        publishable_progress_paths = list(effective_paths)
+                        publishable_age_s = now - publishable_progress_seen_at
+                        if publishable_age_s >= float(no_edit_recheck_s):
+                            publishable_progress_finalized = True
+                            log.info(
+                                "No-edit watchdog observed durable publishable file changes "
+                                f"({_describe_publishable_paths(effective_paths)}) for "
+                                f"{int(publishable_age_s)}s; stopping Codex early so "
+                                "QualityGate/ValidationGate can use the remaining budget."
+                            )
+                            _terminate_active_child()
+                            break
                     no_edit_deadline = now + float(no_edit_recheck_s)
                     log.info(
                         "No-edit watchdog observed publishable-looking file changes "
@@ -2639,6 +2815,35 @@ def _run_codex_task(
                 "cooldownMs": _NO_PUBLISHABLE_FAILURE_COOLDOWN_MS,
             }
 
+        if publishable_progress_finalized:
+            changed_paths, _, effective_paths = _codex_changed_paths(repo, baseline_snapshot)
+            effective_paths = effective_paths or publishable_progress_paths
+            last_message = _read_text_if_exists(last_message_path)
+            log_git_status(repo, log)
+            prefix = (
+                "Codex produced durable publishable file changes. PushPals stopped the "
+                "Codex child early to preserve validation and revision budget; the normal "
+                "QualityGate/ValidationGate will catch any incomplete edit."
+            )
+            return {
+                "ok": True,
+                "summary": (
+                    "openai_codex stopped after durable publishable progress "
+                    f"({len(effective_paths)} file(s))"
+                ),
+                "stdout": _truncate(
+                    _build_success_stdout(
+                        effective_paths=effective_paths,
+                        last_message=last_message,
+                        trace_excerpt=trace_excerpt,
+                        prefix=prefix,
+                    )
+                ),
+                "stderr": _truncate(stderr),
+                "exitCode": 0,
+                "usage": usage,
+            }
+
         if no_edit_watchdog_fired:
             startup_stall = _codex_trace_is_startup_stall(stdout_trace)
             if startup_stall and startup_stall_recovery_attempt < _MAX_STARTUP_STALL_RECOVERY_ATTEMPTS:

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

@@ -1347,6 +1347,259 @@ class OpenAICodexRuntimeConfigTests(unittest.TestCase):
         self.assertIn("Patched immediately after no-edit recovery", str(result.get("stdout") or ""))
         self.assertIn("src/", str(result.get("stdout") or ""))
 
+    def test_run_codex_task_no_edit_watchdog_allows_command_backed_discovery(self) -> None:
+        with tempfile.TemporaryDirectory(prefix="pushpals-codex-no-edit-command-grace-") as temp_dir:
+            repo = Path(temp_dir) / "repo"
+            repo.mkdir(parents=True, exist_ok=True)
+            (repo / "README.md").write_text("# command grace repo\n", encoding="utf-8")
+            subprocess.run(["git", "init"], cwd=repo, check=True, capture_output=True, text=True)
+            subprocess.run(
+                ["git", "config", "user.name", "PushPals Test"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+            subprocess.run(
+                ["git", "config", "user.email", "pushpals-tests@example.com"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+            subprocess.run(["git", "add", "README.md"], cwd=repo, check=True, capture_output=True, text=True)
+            subprocess.run(
+                ["git", "commit", "-m", "chore: seed command grace repo"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+
+            stub_path = Path(temp_dir) / "fake_codex_no_edit_command_grace.py"
+            stub_path.write_text(
+                "\n".join(
+                    [
+                        "from pathlib import Path",
+                        "import json",
+                        "import sys",
+                        "import time",
+                        "",
+                        "argv = sys.argv[1:]",
+                        "last_message_path = None",
+                        "for index, arg in enumerate(argv):",
+                        "    if arg == '--output-last-message' and index + 1 < len(argv):",
+                        "        last_message_path = argv[index + 1]",
+                        "        break",
+                        "",
+                        "sys.stdin.read()",
+                        "print(json.dumps({'type': 'thread.started'}), flush=True)",
+                        "print(json.dumps({'type': 'turn.started'}), flush=True)",
+                        "print(json.dumps({'type': 'item.started', 'item': {'id': 'cmd-read-target', 'type': 'command_execution', 'command': 'sed -n 1,120p README.md', 'status': 'in_progress'}}), flush=True)",
+                        "time.sleep(1.4)",
+                        "print(json.dumps({'type': 'item.completed', 'item': {'id': 'cmd-read-target', 'type': 'command_execution', 'command': 'sed -n 1,120p README.md', 'status': 'completed', 'exit_code': 0, 'aggregated_output': '# command grace repo'}}), flush=True)",
+                        "time.sleep(1.6)",
+                        "Path('src').mkdir(exist_ok=True)",
+                        "Path('src/command-grace.txt').write_text('patched after command-backed discovery\\n', encoding='utf-8')",
+                        "if last_message_path:",
+                        "    Path(last_message_path).write_text('Patched after command-backed discovery.', encoding='utf-8')",
+                        "print(json.dumps({'type': 'item.completed', 'item': {'type': 'message', 'text': 'Patched after command-backed discovery.'}}), flush=True)",
+                    ]
+                ),
+                encoding="utf-8",
+            )
+
+            env_overrides = {
+                "PUSHPALS_OPENAI_CODEX_BIN_JSON": json.dumps([sys.executable, str(stub_path)]),
+                "PUSHPALS_OPENAI_CODEX_AUTH_MODE": "api_key",
+                "OPENAI_API_KEY": "pushpals-no-edit-command-grace-test-key",
+                "WORKERPALS_OPENAI_CODEX_JSON": "true",
+                "WORKERPALS_OPENAI_CODEX_TIMEOUT_S": "20",
+                "WORKERPALS_OPENAI_CODEX_NO_EDIT_WATCHDOG_S": "1",
+                "WORKERPALS_OPENAI_CODEX_NO_EDIT_COMMAND_GRACE_S": "5",
+                "WORKERPALS_OPENAI_CODEX_PROGRESS_LOG_INTERVAL_S": "1",
+            }
+            with mock.patch.dict(os.environ, env_overrides, clear=False):
+                result = _run_codex_task(
+                    str(repo),
+                    "Add one focused contract assertion after inspecting the hinted test.",
+                    [],
+                )
+
+        self.assertTrue(result.get("ok"), result)
+        self.assertEqual(result.get("exitCode"), 0)
+        self.assertIn("Patched after command-backed discovery", str(result.get("stdout") or ""))
+        self.assertIn("src/", str(result.get("stdout") or ""))
+
+    def test_run_codex_task_no_edit_watchdog_extends_after_later_command_progress(self) -> None:
+        with tempfile.TemporaryDirectory(prefix="pushpals-codex-no-edit-late-command-") as temp_dir:
+            repo = Path(temp_dir) / "repo"
+            repo.mkdir(parents=True, exist_ok=True)
+            (repo / "README.md").write_text("# late command grace repo\n", encoding="utf-8")
+            subprocess.run(["git", "init"], cwd=repo, check=True, capture_output=True, text=True)
+            subprocess.run(
+                ["git", "config", "user.name", "PushPals Test"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+            subprocess.run(
+                ["git", "config", "user.email", "pushpals-tests@example.com"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+            subprocess.run(["git", "add", "README.md"], cwd=repo, check=True, capture_output=True, text=True)
+            subprocess.run(
+                ["git", "commit", "-m", "chore: seed late command repo"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+
+            stub_path = Path(temp_dir) / "fake_codex_late_command_grace.py"
+            stub_path.write_text(
+                "\n".join(
+                    [
+                        "from pathlib import Path",
+                        "import json",
+                        "import sys",
+                        "import time",
+                        "",
+                        "argv = sys.argv[1:]",
+                        "last_message_path = None",
+                        "for index, arg in enumerate(argv):",
+                        "    if arg == '--output-last-message' and index + 1 < len(argv):",
+                        "        last_message_path = argv[index + 1]",
+                        "        break",
+                        "",
+                        "sys.stdin.read()",
+                        "print(json.dumps({'type': 'thread.started'}), flush=True)",
+                        "print(json.dumps({'type': 'turn.started'}), flush=True)",
+                        "print(json.dumps({'type': 'item.started', 'item': {'id': 'cmd-one', 'type': 'command_execution', 'command': 'cat README.md', 'status': 'in_progress'}}), flush=True)",
+                        "time.sleep(0.2)",
+                        "print(json.dumps({'type': 'item.completed', 'item': {'id': 'cmd-one', 'type': 'command_execution', 'command': 'cat README.md', 'status': 'completed', 'exit_code': 0}}), flush=True)",
+                        "time.sleep(2.2)",
+                        "print(json.dumps({'type': 'item.started', 'item': {'id': 'cmd-two', 'type': 'command_execution', 'command': 'ls', 'status': 'in_progress'}}), flush=True)",
+                        "time.sleep(0.2)",
+                        "print(json.dumps({'type': 'item.completed', 'item': {'id': 'cmd-two', 'type': 'command_execution', 'command': 'ls', 'status': 'completed', 'exit_code': 0}}), flush=True)",
+                        "time.sleep(2.0)",
+                        "Path('src').mkdir(exist_ok=True)",
+                        "Path('src/late-command-grace.txt').write_text('patched after later command progress\\n', encoding='utf-8')",
+                        "if last_message_path:",
+                        "    Path(last_message_path).write_text('Patched after later command progress.', encoding='utf-8')",
+                        "print(json.dumps({'type': 'item.completed', 'item': {'type': 'message', 'text': 'Patched after later command progress.'}}), flush=True)",
+                    ]
+                ),
+                encoding="utf-8",
+            )
+
+            env_overrides = {
+                "PUSHPALS_OPENAI_CODEX_BIN_JSON": json.dumps([sys.executable, str(stub_path)]),
+                "PUSHPALS_OPENAI_CODEX_AUTH_MODE": "api_key",
+                "OPENAI_API_KEY": "pushpals-no-edit-late-command-test-key",
+                "WORKERPALS_OPENAI_CODEX_JSON": "true",
+                "WORKERPALS_OPENAI_CODEX_TIMEOUT_S": "20",
+                "WORKERPALS_OPENAI_CODEX_NO_EDIT_WATCHDOG_S": "1",
+                "WORKERPALS_OPENAI_CODEX_NO_EDIT_COMMAND_GRACE_S": "3",
+                "WORKERPALS_OPENAI_CODEX_PROGRESS_LOG_INTERVAL_S": "1",
+            }
+            with mock.patch.dict(os.environ, env_overrides, clear=False):
+                result = _run_codex_task(
+                    str(repo),
+                    "Add one focused contract assertion after a later targeted read.",
+                    [],
+                )
+
+        self.assertTrue(result.get("ok"), result)
+        self.assertEqual(result.get("exitCode"), 0)
+        self.assertIn("Patched after later command progress", str(result.get("stdout") or ""))
+        self.assertIn("src/", str(result.get("stdout") or ""))
+
+    def test_run_codex_task_finalizes_after_durable_publishable_progress(self) -> None:
+        with tempfile.TemporaryDirectory(prefix="pushpals-codex-durable-progress-") as temp_dir:
+            repo = Path(temp_dir) / "repo"
+            repo.mkdir(parents=True, exist_ok=True)
+            (repo / "README.md").write_text("# durable progress repo\n", encoding="utf-8")
+            subprocess.run(["git", "init"], cwd=repo, check=True, capture_output=True, text=True)
+            subprocess.run(
+                ["git", "config", "user.name", "PushPals Test"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+            subprocess.run(
+                ["git", "config", "user.email", "pushpals-tests@example.com"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+            subprocess.run(["git", "add", "README.md"], cwd=repo, check=True, capture_output=True, text=True)
+            subprocess.run(
+                ["git", "commit", "-m", "chore: seed durable progress repo"],
+                cwd=repo,
+                check=True,
+                capture_output=True,
+                text=True,
+            )
+
+            stub_path = Path(temp_dir) / "fake_codex_durable_progress.py"
+            stub_path.write_text(
+                "\n".join(
+                    [
+                        "from pathlib import Path",
+                        "import json",
+                        "import sys",
+                        "import time",
+                        "",
+                        "argv = sys.argv[1:]",
+                        "last_message_path = None",
+                        "for index, arg in enumerate(argv):",
+                        "    if arg == '--output-last-message' and index + 1 < len(argv):",
+                        "        last_message_path = argv[index + 1]",
+                        "        break",
+                        "",
+                        "sys.stdin.read()",
+                        "print(json.dumps({'type': 'thread.started'}), flush=True)",
+                        "print(json.dumps({'type': 'turn.started'}), flush=True)",
+                        "Path('src').mkdir(exist_ok=True)",
+                        "Path('src/durable-progress.txt').write_text('durable patch\\n', encoding='utf-8')",
+                        "if last_message_path:",
+                        "    Path(last_message_path).write_text('Created durable patch and kept thinking.', encoding='utf-8')",
+                        "print(json.dumps({'type': 'item.completed', 'item': {'type': 'message', 'text': 'Created durable patch and kept thinking.'}}), flush=True)",
+                        "time.sleep(10)",
+                    ]
+                ),
+                encoding="utf-8",
+            )
+
+            env_overrides = {
+                "PUSHPALS_OPENAI_CODEX_BIN_JSON": json.dumps([sys.executable, str(stub_path)]),
+                "PUSHPALS_OPENAI_CODEX_AUTH_MODE": "api_key",
+                "OPENAI_API_KEY": "pushpals-durable-progress-test-key",
+                "WORKERPALS_OPENAI_CODEX_JSON": "true",
+                "WORKERPALS_OPENAI_CODEX_TIMEOUT_S": "20",
+                "WORKERPALS_OPENAI_CODEX_NO_EDIT_WATCHDOG_S": "1",
+                "WORKERPALS_OPENAI_CODEX_NO_EDIT_RECHECK_S": "1",
+                "WORKERPALS_OPENAI_CODEX_PROGRESS_LOG_INTERVAL_S": "1",
+            }
+            with mock.patch.dict(os.environ, env_overrides, clear=False):
+                result = _run_codex_task(
+                    str(repo),
+                    "Make a focused patch and stop once it is durable.",
+                    [],
+                )
+
+        self.assertTrue(result.get("ok"), result)
+        self.assertEqual(result.get("exitCode"), 0)
+        self.assertIn("stopped after durable publishable progress", str(result.get("summary") or ""))
+        self.assertIn("src/", str(result.get("stdout") or ""))
+
     def test_run_codex_task_recovery_attempt_is_still_guarded_by_no_edit_watchdog(self) -> None:
         with tempfile.TemporaryDirectory(prefix="pushpals-codex-no-edit-watchdog-fail-") as temp_dir:
             repo = Path(temp_dir) / "repo"

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

@@ -335,6 +335,22 @@ export function shouldSkipCriticForDeterministicValidationRevision(opts: {
   return opts.validationRuns.some(isDeterministicFastValidationFailure);
 }
 
+export function shouldSkipCriticToPreserveRevisionBudget(opts: {
+  deterministicRequiresRevision: boolean;
+  remainingBudgetMs: number;
+  minimumRevisionBudgetMs: number;
+  criticTimeoutMs: number;
+  criticTimeoutBehavior: "skip" | "retry_once" | "block" | string;
+}): boolean {
+  if (!opts.deterministicRequiresRevision) return false;
+  const remainingBudgetMs = Math.max(0, Math.floor(opts.remainingBudgetMs));
+  const minimumRevisionBudgetMs = Math.max(0, Math.floor(opts.minimumRevisionBudgetMs));
+  const criticTimeoutMs = Math.max(0, Math.floor(opts.criticTimeoutMs));
+  const criticAttempts = opts.criticTimeoutBehavior === "retry_once" ? 2 : 1;
+  const criticWorstCaseMs = criticTimeoutMs * criticAttempts;
+  return remainingBudgetMs < minimumRevisionBudgetMs + criticWorstCaseMs;
+}
+
 export function workerAttemptRolloutScore(params: {
   executorElapsedMs: number;
   qualityElapsedMs: number;
@@ -7975,11 +7991,23 @@ export async function executeJob(
         validationOutsideTaskScope,
         validationRuns: quality.validationRuns,
       });
+    const preCriticRevisionBudget = qualityRevisionBudgetDecision({
+      jobElapsedMs: Date.now() - jobStartedAt,
+      executionBudgetMs,
+    });
+    const skipCriticForRevisionBudget = shouldSkipCriticToPreserveRevisionBudget({
+      deterministicRequiresRevision: preCriticDeterministicRequiresRevision,
+      remainingBudgetMs: preCriticRevisionBudget.remainingBudgetMs,
+      minimumRevisionBudgetMs: preCriticRevisionBudget.minimumRevisionBudgetMs,
+      criticTimeoutMs: resolveQualityCriticTimeoutMs(runtimeConfig),
+      criticTimeoutBehavior: resolveQualityCriticTimeoutBehavior(runtimeConfig),
+    });
     const critic =
       quality.skipped ||
       !qualityGatePolicy.criticGateEnabled ||
       skipCriticAfterExecutorTimeout ||
-      skipCriticForDeterministicValidationRevision
+      skipCriticForDeterministicValidationRevision ||
+      skipCriticForRevisionBudget
         ? null
         : executor === "openai_codex"
           ? await runCodexCriticReview(repo, attemptParams, qualityForCritic, runtimeConfig, onLog)
@@ -8020,6 +8048,11 @@ export async function executeJob(
         "stdout",
         "[CriticGate] Skipping critic because deterministic fast validation already requires a quality revision.",
       );
+    } else if (skipCriticForRevisionBudget) {
+      onLog?.(
+        "stdout",
+        `[CriticGate] Skipping critic because deterministic quality already requires revision and remaining budget (${preCriticRevisionBudget.remainingBudgetMs}ms) must be reserved for the next worker turn.`,
+      );
     }
     const rolloutScore = workerAttemptRolloutScore({
       executorElapsedMs,