copilot suggestions

Author: qchapp

src/ai_agent/utils/cache_db.py

@@ -27,10 +27,26 @@
 
 log = logging.getLogger("cache_db")
 
-_DEFAULT_DB_PATH = os.getenv(
-    "CACHE_DB_PATH",
-    str(Path(tempfile.gettempdir()) / "ai_agent_cache.db"),
-)
+
+def _default_db_path() -> str:
+    """Return a per-user cache DB path so different OS users cannot share data.
+
+    On POSIX systems the current UID is embedded in the filename, ensuring
+    each user gets an isolated file under the system temp directory.  The
+    caller can override the path entirely via ``CACHE_DB_PATH``.
+    """
+    env = os.getenv("CACHE_DB_PATH")
+    if env:
+        return env
+    uid_suffix = ""
+    try:
+        uid_suffix = f"_{os.getuid()}"
+    except AttributeError:
+        pass  # Windows — no getuid(); fall back to a single shared filename
+    return str(Path(tempfile.gettempdir()) / f"ai_agent_cache{uid_suffix}.db")
+
+
+_DEFAULT_DB_PATH = _default_db_path()
 
 
 class CacheDB:
@@ -52,6 +68,13 @@ def __init__(self, db_path: str | Path | None = None) -> None:
         self._conn = sqlite3.connect(path, check_same_thread=False)
         self._lock = threading.Lock()
 
+        # Restrict the DB file to the owner only so other OS users cannot
+        # read cached data.  The chmod is best-effort (no-op on Windows).
+        if path != ":memory:":
+            try:
+                os.chmod(path, 0o600)
+            except Exception:
+                log.warning("Could not set owner-only permissions on cache DB: %s", path)
         self._conn.execute("PRAGMA journal_mode=WAL")
         self._conn.execute("PRAGMA synchronous=NORMAL")
         self._conn.executescript(

src/ai_agent/utils/image_meta.py

@@ -10,18 +10,33 @@
 from PIL import Image
 import tifffile as tiff
 
-from ai_agent.utils.cache_db import get_cache_db
+from ai_agent.utils.cache_db import CacheDB, get_cache_db
 
 # ---------------------------------------------------------------------------
 # SQLite-backed metadata cache (keyed by resolved-path + mtime + size)
 # Avoids re-reading large files (e.g. TIFF stacks) on every retrieval call.
+#
+# PRIVACY NOTE: DICOM-derived metadata can contain sensitive identifying
+# fields (Study/Series descriptions, institution names, patient context).
+# By default the metadata namespace uses an in-memory-only CacheDB so
+# nothing is written to disk.  Set IMAGE_META_CACHE_PERSIST=1 to opt into
+# on-disk persistence (e.g. for long-running servers with large TIFF stacks).
 # ---------------------------------------------------------------------------
 _META_CACHE_MAX = int(os.getenv("IMAGE_META_CACHE_MAX", "128"))
+_META_CACHE_PERSIST = os.getenv("IMAGE_META_CACHE_PERSIST", "0").strip() == "1"
 
 _META_NS = "meta"
 
 _meta_log = __import__("logging").getLogger("cache_db.meta")
 
+# In-memory-only DB used when persistence is disabled (the default).
+_meta_mem_db: CacheDB | None = None if _META_CACHE_PERSIST else CacheDB(":memory:")
+
+
+def _get_meta_db() -> CacheDB:
+    """Return the CacheDB instance to use for image metadata."""
+    return get_cache_db() if _META_CACHE_PERSIST else _meta_mem_db  # type: ignore[return-value]
+
 
 def _meta_cache_key(p: Path) -> str:
     """Stable cache key derived from resolved path, mtime, and size."""
@@ -34,15 +49,15 @@ def _meta_cache_key(p: Path) -> str:
 
 def _meta_cache_get(key: str) -> Optional[str]:
     try:
-        return get_cache_db().get(_META_NS, key)
+        return _get_meta_db().get(_META_NS, key)
     except Exception:
         _meta_log.warning("Metadata cache get failed; skipping cache.", exc_info=True)
         return None
 
 
 def _meta_cache_set(key: str, value: str) -> None:
     try:
-        get_cache_db().set(_META_NS, key, value, max_entries=_META_CACHE_MAX)
+        _get_meta_db().set(_META_NS, key, value, max_entries=_META_CACHE_MAX)
     except Exception:
         _meta_log.warning("Metadata cache set failed; continuing without caching.", exc_info=True)