fix(session): log skipped unreadable transcripts during search

Surface invalid-UTF-8/corrupt transcript skips via a debug log in both
the JSON scan and indexed candidate loader so a corrupt session is
observable instead of an indistinguishable silent miss. Search stays
read-only (no quarantine/rewrite), preserving corrupt bytes.

Co-authored-by: Mervin Praison <[email protected]>

src/praisonai-agents/praisonaiagents/session/sqlite_store.py

@@ -560,7 +560,16 @@ class SqliteSessionStore(DefaultSessionStore):
 
                 with open(filepath, "r", encoding="utf-8") as f:
                     data = json.load(f)
-            except (json.JSONDecodeError, IOError, OSError):
+            except (UnicodeDecodeError, json.JSONDecodeError, IOError, OSError) as exc:
+                # Read-only path: never quarantine or rewrite during search.
+                # Log so an unreadable indexed candidate is observable rather
+                # than a silent miss, then keep scoring the remaining valid
+                # sessions.
+                logger.debug(
+                    "Skipping unreadable session file %s during search: %s",
+                    filepath,
+                    exc,
+                )
                 continue
 
             messages = self._searchable_messages(data)

src/praisonai-agents/praisonaiagents/session/store.py

@@ -2075,7 +2075,15 @@ class DefaultSessionStore:
             try:
                 with open(filepath, "r", encoding="utf-8") as f:
                     data = json.load(f)
-            except (json.JSONDecodeError, IOError):
+            except (UnicodeDecodeError, json.JSONDecodeError, IOError) as exc:
+                # Read-only path: never quarantine or rewrite during search.
+                # Log so an unreadable transcript is observable rather than a
+                # silent miss, then keep scoring the remaining valid sessions.
+                logger.debug(
+                    "Skipping unreadable session file %s during search: %s",
+                    filepath,
+                    exc,
+                )
                 continue
 
             messages = self._searchable_messages(data)

src/praisonai-agents/tests/unit/session/test_session_search_invalid_utf8.py

@@ -0,0 +1,34 @@
+"""Unreadable transcript bytes must not interrupt recall of valid sessions."""
+
+import pytest
+
+from praisonaiagents.session import SqliteSessionStore
+from praisonaiagents.session.store import DefaultSessionStore
+
+
+@pytest.mark.parametrize("backend", ["json", "fts", "like"])
+def test_search_skips_invalid_utf8_candidate(tmp_path, backend):
+    store = (
+        DefaultSessionStore(session_dir=str(tmp_path))
+        if backend == "json"
+        else SqliteSessionStore(session_dir=str(tmp_path), db_path=":memory:")
+    )
+    try:
+        assert store.add_message("bad", "user", "migration needle")
+        assert store.add_message("valid", "user", "migration needle")
+        if backend == "fts" and not store._fts_available:
+            pytest.skip("SQLite build lacks FTS5")
+        if backend == "like":
+            store._fts_available = False
+        assert {hit.session_id for hit in store.search("needle")} == {"bad", "valid"}
+        bad = tmp_path / "bad.json"
+        payload = b"\xff\xfe"
+        bad.write_bytes(payload)
+        hits = store.search("needle")
+        assert [hit.session_id for hit in hits] == ["valid"]
+        assert hits[0].messages[0]["content"] == "migration needle"
+        assert bad.read_bytes() == payload
+    finally:
+        conn = getattr(store, "_conn", None)
+        if conn is not None:
+            conn.close()