From 2302a5a053d4ce38e856bc1dae9ec0bda7fd7baa Mon Sep 17 00:00:00 2001
From: "Sharad." <heysharad@proton.me>
Date: Wed, 16 Sep 2026 21:01:44 +0530
Subject: [PATCH] fix: CodexParser skips zero-length messages
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Empty-content message items were emitted as empty messages, polluting
token totals — unlike the Claude and OpenAI parsers which already skip
empties. Skip message items whose text is empty or whitespace-only.

Closes #9
---
 src/ctxlens/parsers/codex.py |  2 ++
 tests/test_parsers.py        | 22 ++++++++++++++++++++++
 2 files changed, 24 insertions(+)

diff --git a/src/ctxlens/parsers/codex.py b/src/ctxlens/parsers/codex.py
index 629ad9e..de43208 100644
--- a/src/ctxlens/parsers/codex.py
+++ b/src/ctxlens/parsers/codex.py
@@ -92,6 +92,8 @@ def _emit_item(self, item: dict, turn: int, out: list[Message]) -> int:
         if itype == "message":
             role = item.get("role", "user")
             text = self._content_text(item.get("content"))
+            if not text.strip():
+                return turn
             seg = Segment.ASSISTANT if role == "assistant" else Segment.USER
             if role == "system":
                 seg = Segment.SYSTEM
diff --git a/tests/test_parsers.py b/tests/test_parsers.py
index 06c806b..e732364 100644
--- a/tests/test_parsers.py
+++ b/tests/test_parsers.py
@@ -71,6 +71,28 @@ def test_parse_codex(codex_session):
     assert s.meta.get("session_id") == "codex-abc123"
 
 
+def test_parse_codex_skips_empty_content_messages():
+    raw = json.dumps(
+        {
+            "session": {"id": "codex-empty"},
+            "instructions": "Be concise.",
+            "items": [
+                {"type": "message", "role": "user", "content": "hello"},
+                {"type": "message", "role": "user", "content": ""},
+                {"type": "message", "role": "assistant", "content": "  "},
+                {"type": "message", "role": "assistant", "content": "world"},
+            ],
+        }
+    )
+    s = parse_text(raw, fmt="codex-session")
+    texts = [m.text for m in s.messages]
+    assert "hello" in texts
+    assert "world" in texts
+    assert "" not in texts
+    assert "  " not in texts
+    assert len(texts) == len(["hello", "world"]) + 1  # +1 instructions/session
+
+
 def test_parse_openai_object_has_tool_defs(openai_chat):
     s = parse_file(openai_chat)
     assert any(m.segment == Segment.TOOL_DEFINITIONS for m in s.messages)