fix(harness): retry timed-out Codex steps

tests/harness/test_execute.py

@@ -1,11 +1,14 @@
 import os
+import subprocess
 import sys
 from pathlib import Path
 
 import pytest
 
+import scripts.execute as execute_module
 from scripts.execute import (
     CodexEnvironmentError,
+    StepExecutor,
     build_codex_command,
     build_codex_environment,
     configure_standard_streams,
@@ -105,6 +108,95 @@ def test_run_utf8_process_sends_prompt_as_utf8(tmp_path):
     assert result.stdout == prompt
 
 
+def test_invoke_codex_records_timeout_with_partial_output(tmp_path, monkeypatch):
+    phase_dir = tmp_path / "phases" / "test-phase"
+    phase_dir.mkdir(parents=True)
+    (phase_dir / "step0.md").write_text("# Timed step\n", encoding="utf-8")
+    codex = tmp_path / "codex.exe"
+    codex.touch()
+
+    executor = StepExecutor.__new__(StepExecutor)
+    executor._root = str(tmp_path)
+    executor._phase_dir = phase_dir
+    executor._codex_add_dirs = ()
+
+    monkeypatch.setattr(
+        execute_module.shutil, "which", lambda executable: str(codex)
+    )
+
+    def raise_timeout(*args, **kwargs):
+        raise subprocess.TimeoutExpired(
+            cmd=["codex", "exec"],
+            timeout=1800,
+            output=b"partial stdout",
+            stderr=b"partial stderr",
+        )
+
+    monkeypatch.setattr(execute_module, "run_utf8_process", raise_timeout)
+
+    output = executor._invoke_codex(
+        {"step": 0, "name": "timed-step"}, "preamble\n"
+    )
+
+    assert output["exitCode"] == 124
+    assert output["stdout"] == "partial stdout"
+    assert "partial stderr" in output["stderr"]
+    assert "timed out after 1800 seconds" in output["stderr"]
+    assert (
+        execute_module.json.loads(
+            (phase_dir / "step0-output.json").read_text(encoding="utf-8")
+        )
+        == output
+    )
+
+
+def test_execute_single_step_retries_timeout_three_times(tmp_path, monkeypatch):
+    index_file = tmp_path / "index.json"
+    index_file.write_text(
+        execute_module.json.dumps(
+            {
+                "steps": [
+                    {"step": 0, "name": "timed-step", "status": "pending"}
+                ]
+            }
+        ),
+        encoding="utf-8",
+    )
+
+    executor = StepExecutor.__new__(StepExecutor)
+    executor._index_file = index_file
+    executor._total = 1
+    executor._project = "FESA"
+    executor._phase_dir_name = "test-phase"
+    attempts = 0
+
+    def timeout_output(step, preamble):
+        nonlocal attempts
+        attempts += 1
+        return {
+            "step": 0,
+            "name": "timed-step",
+            "exitCode": 124,
+            "stdout": "",
+            "stderr": "Codex timed out after 1800 seconds.",
+        }
+
+    monkeypatch.setattr(executor, "_invoke_codex", timeout_output)
+    monkeypatch.setattr(executor, "_commit_step", lambda *args: None)
+    monkeypatch.setattr(executor, "_update_top_index", lambda *args: None)
+
+    with pytest.raises(SystemExit) as exit_info:
+        executor._execute_single_step(
+            {"step": 0, "name": "timed-step", "status": "pending"}, ""
+        )
+
+    assert exit_info.value.code == 1
+    assert attempts == 3
+    index = execute_module.json.loads(index_file.read_text(encoding="utf-8"))
+    assert index["steps"][0]["status"] == "error"
+    assert "timed out after 1800 seconds" in index["steps"][0]["error_message"]
+
+
 def test_configure_standard_streams_replaces_unencodable_status_characters():
     class RecordingStream:
         def __init__(self):