diff --git a/batch_runner.py b/batch_runner.py
index f1f4f464f..b95a5cc82 100644
--- a/batch_runner.py
+++ b/batch_runner.py
@@ -29,8 +29,6 @@ from typing import List, Dict, Any, Optional, Tuple
 from datetime import datetime
 from multiprocessing import Pool, Lock
 import traceback
-import tempfile
-
 from rich.progress import Progress, SpinnerColumn, BarColumn, TextColumn, TimeRemainingColumn, MofNCompleteColumn
 from rich.console import Console
 import fire
@@ -703,32 +701,12 @@ class BatchRunner:
         """
         checkpoint_data["last_updated"] = datetime.now().isoformat()
 
-        def _atomic_write():
-            """Write checkpoint atomically (temp file + replace) to avoid corruption on crash."""
-            self.checkpoint_file.parent.mkdir(parents=True, exist_ok=True)
-            fd, tmp_path = tempfile.mkstemp(
-                dir=str(self.checkpoint_file.parent),
-                prefix='.checkpoint_',
-                suffix='.tmp',
-            )
-            try:
-                with os.fdopen(fd, 'w', encoding='utf-8') as f:
-                    json.dump(checkpoint_data, f, indent=2, ensure_ascii=False)
-                    f.flush()
-                    os.fsync(f.fileno())
-                os.replace(tmp_path, self.checkpoint_file)
-            except BaseException:
-                try:
-                    os.unlink(tmp_path)
-                except OSError:
-                    pass
-                raise
-
+        from utils import atomic_json_write
         if lock:
             with lock:
-                _atomic_write()
+                atomic_json_write(self.checkpoint_file, checkpoint_data)
         else:
-            _atomic_write()
+            atomic_json_write(self.checkpoint_file, checkpoint_data)
     
     def _scan_completed_prompts_by_content(self) -> set:
         """
diff --git a/tests/test_atomic_json_write.py b/tests/test_atomic_json_write.py
new file mode 100644
index 000000000..681b7d8a8
--- /dev/null
+++ b/tests/test_atomic_json_write.py
@@ -0,0 +1,132 @@
+"""Tests for utils.atomic_json_write — crash-safe JSON file writes."""
+
+import json
+import os
+from pathlib import Path
+from unittest.mock import patch
+
+import pytest
+
+from utils import atomic_json_write
+
+
+class TestAtomicJsonWrite:
+    """Core atomic write behavior."""
+
+    def test_writes_valid_json(self, tmp_path):
+        target = tmp_path / "data.json"
+        data = {"key": "value", "nested": {"a": 1}}
+        atomic_json_write(target, data)
+
+        result = json.loads(target.read_text(encoding="utf-8"))
+        assert result == data
+
+    def test_creates_parent_directories(self, tmp_path):
+        target = tmp_path / "deep" / "nested" / "dir" / "data.json"
+        atomic_json_write(target, {"ok": True})
+
+        assert target.exists()
+        assert json.loads(target.read_text())["ok"] is True
+
+    def test_overwrites_existing_file(self, tmp_path):
+        target = tmp_path / "data.json"
+        target.write_text('{"old": true}')
+
+        atomic_json_write(target, {"new": True})
+        result = json.loads(target.read_text())
+        assert result == {"new": True}
+
+    def test_preserves_original_on_serialization_error(self, tmp_path):
+        target = tmp_path / "data.json"
+        original = {"preserved": True}
+        target.write_text(json.dumps(original))
+
+        # Try to write non-serializable data — should fail
+        with pytest.raises(TypeError):
+            atomic_json_write(target, {"bad": object()})
+
+        # Original file should be untouched
+        result = json.loads(target.read_text())
+        assert result == original
+
+    def test_no_leftover_temp_files_on_success(self, tmp_path):
+        target = tmp_path / "data.json"
+        atomic_json_write(target, [1, 2, 3])
+
+        # No .tmp files should be left behind
+        tmp_files = [f for f in tmp_path.iterdir() if ".tmp" in f.name]
+        assert len(tmp_files) == 0
+        assert target.exists()
+
+    def test_no_leftover_temp_files_on_failure(self, tmp_path):
+        target = tmp_path / "data.json"
+
+        with pytest.raises(TypeError):
+            atomic_json_write(target, {"bad": object()})
+
+        # No temp files should be left behind
+        tmp_files = [f for f in tmp_path.iterdir() if ".tmp" in f.name]
+        assert len(tmp_files) == 0
+
+    def test_accepts_string_path(self, tmp_path):
+        target = str(tmp_path / "string_path.json")
+        atomic_json_write(target, {"string": True})
+
+        result = json.loads(Path(target).read_text())
+        assert result == {"string": True}
+
+    def test_writes_list_data(self, tmp_path):
+        target = tmp_path / "list.json"
+        data = [1, "two", {"three": 3}]
+        atomic_json_write(target, data)
+
+        result = json.loads(target.read_text())
+        assert result == data
+
+    def test_empty_list(self, tmp_path):
+        target = tmp_path / "empty.json"
+        atomic_json_write(target, [])
+
+        result = json.loads(target.read_text())
+        assert result == []
+
+    def test_custom_indent(self, tmp_path):
+        target = tmp_path / "custom.json"
+        atomic_json_write(target, {"a": 1}, indent=4)
+
+        text = target.read_text()
+        assert '    "a"' in text  # 4-space indent
+
+    def test_unicode_content(self, tmp_path):
+        target = tmp_path / "unicode.json"
+        data = {"emoji": "🎉", "japanese": "日本語"}
+        atomic_json_write(target, data)
+
+        result = json.loads(target.read_text(encoding="utf-8"))
+        assert result["emoji"] == "🎉"
+        assert result["japanese"] == "日本語"
+
+    def test_concurrent_writes_dont_corrupt(self, tmp_path):
+        """Multiple rapid writes should each produce valid JSON."""
+        import threading
+
+        target = tmp_path / "concurrent.json"
+        errors = []
+
+        def writer(n):
+            try:
+                atomic_json_write(target, {"writer": n, "data": list(range(100))})
+            except Exception as e:
+                errors.append(e)
+
+        threads = [threading.Thread(target=writer, args=(i,)) for i in range(10)]
+        for t in threads:
+            t.start()
+        for t in threads:
+            t.join()
+
+        assert not errors
+        # File should contain valid JSON from one of the writers
+        result = json.loads(target.read_text())
+        assert "writer" in result
+        assert len(result["data"]) == 100
diff --git a/tests/test_batch_runner_checkpoint.py b/tests/test_batch_runner_checkpoint.py
new file mode 100644
index 000000000..ebf9bce7e
--- /dev/null
+++ b/tests/test_batch_runner_checkpoint.py
@@ -0,0 +1,159 @@
+"""Tests for batch_runner checkpoint behavior — incremental writes, resume, atomicity."""
+
+import json
+import os
+from pathlib import Path
+from multiprocessing import Lock
+from unittest.mock import patch, MagicMock
+
+import pytest
+
+# batch_runner uses relative imports, ensure project root is on path
+import sys
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+from batch_runner import BatchRunner
+
+
+@pytest.fixture
+def runner(tmp_path):
+    """Create a BatchRunner with all paths pointing at tmp_path."""
+    prompts_file = tmp_path / "prompts.jsonl"
+    prompts_file.write_text("")
+    output_file = tmp_path / "output.jsonl"
+    checkpoint_file = tmp_path / "checkpoint.json"
+    r = BatchRunner.__new__(BatchRunner)
+    r.run_name = "test_run"
+    r.checkpoint_file = checkpoint_file
+    r.output_file = output_file
+    r.prompts_file = prompts_file
+    return r
+
+
+class TestSaveCheckpoint:
+    """Verify _save_checkpoint writes valid, atomic JSON."""
+
+    def test_writes_valid_json(self, runner):
+        data = {"run_name": "test", "completed_prompts": [1, 2, 3], "batch_stats": {}}
+        runner._save_checkpoint(data)
+
+        result = json.loads(runner.checkpoint_file.read_text())
+        assert result["run_name"] == "test"
+        assert result["completed_prompts"] == [1, 2, 3]
+
+    def test_adds_last_updated(self, runner):
+        data = {"run_name": "test", "completed_prompts": []}
+        runner._save_checkpoint(data)
+
+        result = json.loads(runner.checkpoint_file.read_text())
+        assert "last_updated" in result
+        assert result["last_updated"] is not None
+
+    def test_overwrites_previous_checkpoint(self, runner):
+        runner._save_checkpoint({"run_name": "test", "completed_prompts": [1]})
+        runner._save_checkpoint({"run_name": "test", "completed_prompts": [1, 2, 3]})
+
+        result = json.loads(runner.checkpoint_file.read_text())
+        assert result["completed_prompts"] == [1, 2, 3]
+
+    def test_with_lock(self, runner):
+        lock = Lock()
+        data = {"run_name": "test", "completed_prompts": [42]}
+        runner._save_checkpoint(data, lock=lock)
+
+        result = json.loads(runner.checkpoint_file.read_text())
+        assert result["completed_prompts"] == [42]
+
+    def test_without_lock(self, runner):
+        data = {"run_name": "test", "completed_prompts": [99]}
+        runner._save_checkpoint(data, lock=None)
+
+        result = json.loads(runner.checkpoint_file.read_text())
+        assert result["completed_prompts"] == [99]
+
+    def test_creates_parent_dirs(self, tmp_path):
+        runner_deep = BatchRunner.__new__(BatchRunner)
+        runner_deep.checkpoint_file = tmp_path / "deep" / "nested" / "checkpoint.json"
+
+        data = {"run_name": "test", "completed_prompts": []}
+        runner_deep._save_checkpoint(data)
+
+        assert runner_deep.checkpoint_file.exists()
+
+    def test_no_temp_files_left(self, runner):
+        runner._save_checkpoint({"run_name": "test", "completed_prompts": []})
+
+        tmp_files = [f for f in runner.checkpoint_file.parent.iterdir()
+                     if ".tmp" in f.name]
+        assert len(tmp_files) == 0
+
+
+class TestLoadCheckpoint:
+    """Verify _load_checkpoint reads existing data or returns defaults."""
+
+    def test_returns_empty_when_no_file(self, runner):
+        result = runner._load_checkpoint()
+        assert result.get("completed_prompts", []) == []
+
+    def test_loads_existing_checkpoint(self, runner):
+        data = {"run_name": "test_run", "completed_prompts": [5, 10, 15],
+                "batch_stats": {"0": {"processed": 3}}}
+        runner.checkpoint_file.write_text(json.dumps(data))
+
+        result = runner._load_checkpoint()
+        assert result["completed_prompts"] == [5, 10, 15]
+        assert result["batch_stats"]["0"]["processed"] == 3
+
+    def test_handles_corrupt_json(self, runner):
+        runner.checkpoint_file.write_text("{broken json!!")
+
+        result = runner._load_checkpoint()
+        # Should return empty/default, not crash
+        assert isinstance(result, dict)
+
+
+class TestResumePreservesProgress:
+    """Verify that initializing a run with resume=True loads prior checkpoint."""
+
+    def test_completed_prompts_loaded_from_checkpoint(self, runner):
+        # Simulate a prior run that completed prompts 0-4
+        prior = {
+            "run_name": "test_run",
+            "completed_prompts": [0, 1, 2, 3, 4],
+            "batch_stats": {"0": {"processed": 5}},
+            "last_updated": "2026-01-01T00:00:00",
+        }
+        runner.checkpoint_file.write_text(json.dumps(prior))
+
+        # Load checkpoint like run() does
+        checkpoint_data = runner._load_checkpoint()
+        if checkpoint_data.get("run_name") != runner.run_name:
+            checkpoint_data = {
+                "run_name": runner.run_name,
+                "completed_prompts": [],
+                "batch_stats": {},
+                "last_updated": None,
+            }
+
+        completed_set = set(checkpoint_data.get("completed_prompts", []))
+        assert completed_set == {0, 1, 2, 3, 4}
+
+    def test_different_run_name_starts_fresh(self, runner):
+        prior = {
+            "run_name": "different_run",
+            "completed_prompts": [0, 1, 2],
+            "batch_stats": {},
+        }
+        runner.checkpoint_file.write_text(json.dumps(prior))
+
+        checkpoint_data = runner._load_checkpoint()
+        if checkpoint_data.get("run_name") != runner.run_name:
+            checkpoint_data = {
+                "run_name": runner.run_name,
+                "completed_prompts": [],
+                "batch_stats": {},
+                "last_updated": None,
+            }
+
+        assert checkpoint_data["completed_prompts"] == []
+        assert checkpoint_data["run_name"] == "test_run"
diff --git a/tools/process_registry.py b/tools/process_registry.py
index 0d2f88c42..948f2a4f3 100644
--- a/tools/process_registry.py
+++ b/tools/process_registry.py
@@ -37,7 +37,6 @@ import shlex
 import shutil
 import signal
 import subprocess
-import tempfile
 import threading
 import time
 import uuid
@@ -707,25 +706,9 @@ class ProcessRegistry:
                             "session_key": s.session_key,
                         })
             
-            # Atomic write: temp file + os.replace to avoid corruption on crash
-            CHECKPOINT_PATH.parent.mkdir(parents=True, exist_ok=True)
-            fd, tmp_path = tempfile.mkstemp(
-                dir=str(CHECKPOINT_PATH.parent),
-                prefix='.checkpoint_',
-                suffix='.tmp',
-            )
-            try:
-                with os.fdopen(fd, 'w', encoding='utf-8') as f:
-                    json.dump(entries, f, indent=2, ensure_ascii=False)
-                    f.flush()
-                    os.fsync(f.fileno())
-                os.replace(tmp_path, CHECKPOINT_PATH)
-            except BaseException:
-                try:
-                    os.unlink(tmp_path)
-                except OSError:
-                    pass
-                raise
+            # Atomic write to avoid corruption on crash
+            from utils import atomic_json_write
+            atomic_json_write(CHECKPOINT_PATH, entries)
         except Exception as e:
             logger.debug("Failed to write checkpoint file: %s", e, exc_info=True)
 
@@ -774,26 +757,9 @@ class ProcessRegistry:
                 logger.info("Recovered detached process: %s (pid=%d)", session.command[:60], pid)
 
         # Clear the checkpoint (will be rewritten as processes finish)
-        # Use atomic write to avoid corruption
         try:
-            CHECKPOINT_PATH.parent.mkdir(parents=True, exist_ok=True)
-            fd, tmp_path = tempfile.mkstemp(
-                dir=str(CHECKPOINT_PATH.parent),
-                prefix='.checkpoint_',
-                suffix='.tmp',
-            )
-            try:
-                with os.fdopen(fd, 'w', encoding='utf-8') as f:
-                    f.write("[]")
-                    f.flush()
-                    os.fsync(f.fileno())
-                os.replace(tmp_path, CHECKPOINT_PATH)
-            except BaseException:
-                try:
-                    os.unlink(tmp_path)
-                except OSError:
-                    pass
-                raise
+            from utils import atomic_json_write
+            atomic_json_write(CHECKPOINT_PATH, [])
         except Exception as e:
             logger.debug("Could not clear checkpoint file: %s", e, exc_info=True)
 
diff --git a/utils.py b/utils.py
new file mode 100644
index 000000000..9c8b5e8c6
--- /dev/null
+++ b/utils.py
@@ -0,0 +1,41 @@
+"""Shared utility functions for hermes-agent."""
+
+import json
+import os
+import tempfile
+from pathlib import Path
+from typing import Any, Union
+
+
+def atomic_json_write(path: Union[str, Path], data: Any, *, indent: int = 2) -> None:
+    """Write JSON data to a file atomically.
+
+    Uses temp file + fsync + os.replace to ensure the target file is never
+    left in a partially-written state.  If the process crashes mid-write,
+    the previous version of the file remains intact.
+
+    Args:
+        path: Target file path (will be created or overwritten).
+        data: JSON-serializable data to write.
+        indent: JSON indentation (default 2).
+    """
+    path = Path(path)
+    path.parent.mkdir(parents=True, exist_ok=True)
+
+    fd, tmp_path = tempfile.mkstemp(
+        dir=str(path.parent),
+        prefix=f".{path.stem}_",
+        suffix=".tmp",
+    )
+    try:
+        with os.fdopen(fd, "w", encoding="utf-8") as f:
+            json.dump(data, f, indent=indent, ensure_ascii=False)
+            f.flush()
+            os.fsync(f.fileno())
+        os.replace(tmp_path, path)
+    except BaseException:
+        try:
+            os.unlink(tmp_path)
+        except OSError:
+            pass
+        raise