diff --git a/CLAUDE.md b/CLAUDE.md
new file mode 100644
index 0000000..4dfac01
--- /dev/null
+++ b/CLAUDE.md
@@ -0,0 +1,49 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+A Python utility that synchronizes SharePoint Online folders to local storage using the Microsoft Graph API. Offers both a CLI (`download_sharepoint.py`) and a modern GUI (`sharepoint_gui.py`).
+
+## Running the Tool
+
+```bash
+# Install dependencies
+pip install -r requirements.txt
+
+# GUI mode (recommended for interactive use)
+python sharepoint_gui.py
+
+# CLI mode (for automation/scripting)
+python download_sharepoint.py
+```
+
+Configuration is read from `connection_info.txt` (gitignored — copy from `connection_info.template.txt` and fill in credentials).
+
+## Architecture
+
+Two-file structure with clear separation of concerns:
+
+**`download_sharepoint.py`** — Core engine with four logical layers:
+1. **Authentication** — MSAL `ConfidentialClientApplication` using OAuth 2.0 Client Credentials flow. Tokens are refreshed via `force_refresh=True` when a 401 is received.
+2. **Graph API navigation** — `get_site_id()` → `get_drive_id()` → `process_item_list()` (recursive, handles `@odata.nextLink` pagination).
+3. **Download & resilience** — `download_single_file()` with Range header support for resumable downloads. `get_fresh_download_url()` handles expired pre-signed URLs. The `@retry_request` decorator provides exponential backoff (up to 5 retries, 2^n seconds) for 429s and network errors.
+4. **Concurrency** — `ThreadPoolExecutor` (max 5 workers). A `report_lock` guards the shared error list. A `stop_event` allows the GUI stop button to cancel in-flight work.
+
+**`sharepoint_gui.py`** — CustomTkinter wrapper that:
+- Persists settings to a local JSON file
+- Spawns the core engine in a background thread
+- Patches `requests.get` to route through the GUI's log display
+- Provides a folder browser for `LOCAL_PATH`
+
+## Key Behaviors to Preserve
+
+- **Self-healing sessions**: On 401, the code refreshes both the MSAL access token *and* the pre-signed Graph download URL before retrying — these are two separate expiry mechanisms.
+- **Resumable downloads**: Files are downloaded in 1 MB chunks using HTTP Range headers. Existing files are skipped if their size matches; partial files are resumed from the last byte.
+- **Stop signal**: `stop_event.is_set()` is checked in the download loop and recursive traversal — any new code that loops must respect this.
+
+## Output
+
+- `sharepoint_download.log` — Full operation log
+- `download_report_YYYYMMDD_HHMMSS.csv` — Per-run error report (gitignored)
diff --git a/download_sharepoint.py b/download_sharepoint.py
index 754f249..a00f475 100644
--- a/download_sharepoint.py
+++ b/download_sharepoint.py
@@ -91,7 +91,7 @@ def get_headers(app, force_refresh=False):
     
     if force_refresh or not result or "access_token" not in result:
         logger.info("Refreshing Access Token...")
-        result = app.acquire_token_for_client(scopes=scopes)
+        result = app.acquire_token_for_client(scopes=scopes, force_refresh=True)
         
     if "access_token" in result:
         return {'Authorization': f'Bearer {result["access_token"]}'}
@@ -182,7 +182,7 @@ def download_single_file(app, drive_id, item_id, local_path, expected_size, disp
             if not download_url:
                 return False, f"Failed to refresh download URL: {err}"
             # Retry download with new URL
-            response = requests.get(download_url, headers=resume_header, stream=True, timeout=120)
+            response = safe_get(download_url, resume_header, stream=True, timeout=120)
             
         response.raise_for_status()
         
@@ -291,8 +291,9 @@ def main():
         
         report_file = f"download_report_{datetime.now().strftime('%Y%m%d_%H%M%S')}.csv"
         with open(report_file, 'w', newline='', encoding='utf-8') as f:
-            csv.DictWriter(f, fieldnames=["Path", "Error", "Timestamp"]).writeheader()
-            csv.DictWriter(f, fieldnames=["Path", "Error", "Timestamp"]).writerows(report)
+            writer = csv.DictWriter(f, fieldnames=["Path", "Error", "Timestamp"])
+            writer.writeheader()
+            writer.writerows(report)
         
         logger.info(f"Sync complete. Errors: {len(report)}. Report: {report_file}")