From 8bc5a6872f8503da3413d3b7a52419a7817fdc91 Mon Sep 17 00:00:00 2001
From: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
Date: Wed, 28 Oct 2026 15:42:58 +0800
Subject: [PATCH] fix(benchmark): reject invalid replan capture paths before
 setup

Signed-off-by: LoopX Agent <337587101+loopx-agent@users.noreply.github.com>
---
 benchmark/runtime/harbor.py                  |  6 +++++-
 benchmark/tests/test_shared_codex_runtime.py | 21 ++++++++++++++++++++
 2 files changed, 26 insertions(+), 1 deletion(-)

diff --git a/benchmark/runtime/harbor.py b/benchmark/runtime/harbor.py
index 1dd183ae83..7e4330d93b 100644
--- a/benchmark/runtime/harbor.py
+++ b/benchmark/runtime/harbor.py
@@ -7,7 +7,7 @@
 import shlex
 import tempfile
 import time
-from pathlib import Path
+from pathlib import Path, PurePosixPath
 from typing import Iterable
 
 from harbor.agents.installed.base import with_prompt_template
@@ -147,6 +147,10 @@ def __init__(
                 raise ValueError("checkpoint_request requires its staged guest path")
             from .checkpoint import parse_checkpoint_request
             self._checkpoint_declaration = parse_checkpoint_request(checkpoint_request)
+            if self._checkpoint_declaration.get("trigger") == "fresh_replan":
+                expected = PurePosixPath(_WAKE_LOG_DIR).parent / "checkpoints"
+                if PurePosixPath(self._checkpoint_declaration["destination"]) != expected:
+                    raise ValueError(f"fresh_replan destination must be {expected} in the worker")
         if checkpoint_request_path is not None:
             if self.execution.mode not in {"heartbeat", "turn"}:
                 raise ValueError("checkpoint_request_path requires a settled-work worker")
diff --git a/benchmark/tests/test_shared_codex_runtime.py b/benchmark/tests/test_shared_codex_runtime.py
index 73274ad7c3..c3fb51c2d1 100644
--- a/benchmark/tests/test_shared_codex_runtime.py
+++ b/benchmark/tests/test_shared_codex_runtime.py
@@ -632,6 +632,26 @@ def test_checkpoint_request_is_explicit_and_reaches_the_existing_worker_env(tmp_
             BenchmarkCodex(logs_dir=tmp_path / "invalid", model_name="openai/fixture", **settings)
 
 
+@pytest.mark.parametrize("destination", [
+    "/opt/loopx-benchmark/checkpoints", "/logs/checkpoints", "relative/checkpoints",
+])
+def test_fresh_replan_destination_rejected_before_native_setup(tmp_path, monkeypatch, destination):
+    pytest.importorskip("harbor")
+    from benchmark.runtime.harbor import BenchmarkCodex, CodexOffline
+
+    def unexpected_setup(*args, **kwargs):
+        pytest.fail("Invalid capture configuration reached native setup")
+
+    monkeypatch.setattr(CodexOffline, "__init__", unexpected_setup)
+    request = {"destination": destination,
+        "components": [{"id": "workspace", "source_path": "/task"}],
+        "point_metadata": {}, "trigger": "fresh_replan"}
+    with pytest.raises(ValueError, match="fresh_replan destination"):
+        BenchmarkCodex(logs_dir=tmp_path, model_name="fixture",
+            checkpoint_request_path="/control/request.json",
+            checkpoint_request=json.dumps(request).encode())
+
+
 @pytest.mark.parametrize("existing", [False, True])
 @pytest.mark.parametrize("settings,field,value", [
     ({}, "replan_after_effective_turns", 6),
@@ -806,6 +826,7 @@ def test_harbor_install_receipt_keeps_capture_independent_from_display(tmp_path,
     from benchmark.runtime.checkpoint_wake import CheckpointWake
     wake_root = tmp_path / "wakes"
     wake_root.mkdir()
+    monkeypatch.setattr("benchmark.runtime.harbor._WAKE_LOG_DIR", str(wake_root))
     request = tmp_path / "request.json"
     checkpoint_args = {}
     if trigger is not None: