fix(normalize): discard user/gemini turns before session_metadata sentinel

Agent-Logs-Url: https://github.com/MemPalace/mempalace/sessions/4511e9aa-38e7-440e-a6f8-eda91e576f0f

Co-authored-by: igorls <4753812+igorls@users.noreply.github.com>

mempalace/normalize.py

@@ -296,9 +296,12 @@ def _try_gemini_jsonl(content: str) -> Optional[str]:
 
     Detection requires a ``session_metadata`` record so this parser does
     not false-positive against Claude Code or Codex JSONL passed through
-    the dispatch chain. ``message_update`` entries are skipped — they
+    the dispatch chain. Any ``user``/``gemini`` lines that appear before
-    have no message text. Multiple text blocks within a single message's
+    ``session_metadata`` are discarded — they are treated as preamble
-    content array are concatenated in order, separated by newlines.
+    noise, not conversational turns. ``message_update`` entries are
+    skipped — they have no message text. Multiple text blocks within a
+    single message's content array are concatenated in order, separated
+    by newlines.
     """
     lines = [line.strip() for line in content.strip().split("\n") if line.strip()]
     messages = []
@@ -316,6 +319,11 @@ def _try_gemini_jsonl(content: str) -> Optional[str]:
             has_session_metadata = True
             continue
 
+        # Discard everything (including user/gemini turns) until the
+        # session_metadata sentinel has been seen.
+        if not has_session_metadata:
+            continue
+
         if entry_type not in ("user", "gemini"):
             # Skips message_update, system events, anything else.
             continue

tests/test_normalize.py

@@ -594,6 +594,25 @@ def test_gemini_jsonl_does_not_match_codex():
     assert result is None
 
 
+def test_gemini_jsonl_messages_before_session_metadata_discarded():
+    """user/gemini turns that appear before the session_metadata sentinel must
+    be silently discarded, not counted as conversational messages.  Only turns
+    after the sentinel contribute to the transcript."""
+    lines = [
+        json.dumps({"type": "user", "content": [{"text": "preamble Q"}]}),
+        json.dumps({"type": "gemini", "content": [{"text": "preamble A"}]}),
+        json.dumps({"type": "session_metadata", "sessionId": "s"}),
+        json.dumps({"type": "user", "content": [{"text": "real Q"}]}),
+        json.dumps({"type": "gemini", "content": [{"text": "real A"}]}),
+    ]
+    result = _try_gemini_jsonl("\n".join(lines))
+    assert result is not None
+    assert "preamble Q" not in result
+    assert "preamble A" not in result
+    assert "> real Q" in result
+    assert "real A" in result
+
+
 # ── _try_claude_ai_json ───────────────────────────────────────────────