coze_lab 0.1.42 → 0.1.43

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "coze_lab",
-  "version": "0.1.42",
+  "version": "0.1.43",
   "description": "Configure local AI agents (Claude Code, Codex, OpenClaw) to report traces to CozeLoop",
   "keywords": [
     "cozeloop",

package/scripts/claude-code/cozeloop_hook.py

@@ -550,6 +550,19 @@ def read_new_messages(file_path: str, start_line: int = 0) -> List[Dict[str, Any
 
 # --- Content Helpers ---
 
+def _usage_int(usage: Any, key: str) -> int:
+    """Read a token count from a usage dict, treating missing/None/非数字 一律为 0。
+
+    Claude Code transcript 里 usage 的 cache_* 等字段常【存在但值为 null】，dict.get(key, 0)
+    对显式 null 返回 None 而非 0，后续 None + int / None > 0 会抛 TypeError，导致整个实时
+    上报失败、trace 查不到。这里统一兜底。
+    """
+    if not isinstance(usage, dict):
+        return 0
+    v = usage.get(key)
+    return v if isinstance(v, int) else 0
+
+
 def is_empty_content(content: Any) -> bool:
     """Return True if content carries no meaningful data."""
     if content is None:
@@ -676,7 +689,7 @@ def _group_subagent_steps(progress_msgs: List[Dict[str, Any]]) -> List[Dict[str,
                     existing.extend(content)
                 last_step["tool_calls"].extend(tool_calls)
                 usage = pmsg.get("usage", {})
-                if usage.get("input_tokens", 0) > 0 or usage.get("output_tokens", 0) > 0:
+                if _usage_int(usage, "input_tokens") > 0 or _usage_int(usage, "output_tokens") > 0:
                     last_step["assistant_message"]["message"]["usage"] = usage
             else:
                 steps.append({
@@ -849,7 +862,7 @@ def group_messages_into_turns(messages: List[Dict[str, Any]]) -> List[Dict[str,
                     last_step["tool_calls"].extend(tool_calls)
                     # Carry over usage from the later line (earlier line typically has zeros)
                     usage = message.get("usage", {})
-                    if usage.get("input_tokens", 0) > 0 or usage.get("output_tokens", 0) > 0:
+                    if _usage_int(usage, "input_tokens") > 0 or _usage_int(usage, "output_tokens") > 0:
                         last_step["assistant_message"]["message"]["usage"] = usage
                 else:
                     # New API response — create a new step
@@ -1329,10 +1342,10 @@ def send_turns_to_cozeloop(turns: List[Dict[str, Any]], session_id: str, history
 
                                 # Set token usage for this specific model call
                                 usage = assistant_message_obj.get("usage", {})
-                                input_tokens = usage.get("input_tokens", 0)
-                                output_tokens = usage.get("output_tokens", 0)
-                                cache_creation = usage.get("cache_creation_input_tokens", 0)
-                                cache_read = usage.get("cache_read_input_tokens", 0)
+                                input_tokens = _usage_int(usage, "input_tokens")
+                                output_tokens = _usage_int(usage, "output_tokens")
+                                cache_creation = _usage_int(usage, "cache_creation_input_tokens")
+                                cache_read = _usage_int(usage, "cache_read_input_tokens")
                                 if input_tokens > 0 or cache_creation > 0 or cache_read > 0:
                                     model_span.set_input_tokens(input_tokens + cache_creation + cache_read)
                                 if output_tokens > 0:
@@ -1402,10 +1415,10 @@ def send_turns_to_cozeloop(turns: List[Dict[str, Any]], session_id: str, history
 
                                         # Distribute total usage evenly across sub-agent model steps.
                                         total_usage = tool_call.get("_total_usage", {})
-                                        total_in = (total_usage.get("input_tokens", 0)
-                                                    + total_usage.get("cache_creation_input_tokens", 0)
-                                                    + total_usage.get("cache_read_input_tokens", 0))
-                                        total_out = total_usage.get("output_tokens", 0)
+                                        total_in = (_usage_int(total_usage, "input_tokens")
+                                                    + _usage_int(total_usage, "cache_creation_input_tokens")
+                                                    + _usage_int(total_usage, "cache_read_input_tokens"))
+                                        total_out = _usage_int(total_usage, "output_tokens")
                                         n_model_steps = len(sub_steps)
                                         per_step_in = total_in // n_model_steps if n_model_steps > 0 else 0
                                         per_step_out = total_out // n_model_steps if n_model_steps > 0 else 0
@@ -1778,12 +1791,12 @@ def send_steps_realtime(turns, session_id, history_turns, state, coze_tags_overr
             if text_parts:
                 mspan.set_output("\n".join(text_parts))
             usage = amo.get("usage", {})
-            it_tok = usage.get("input_tokens", 0) + usage.get("cache_creation_input_tokens", 0) + usage.get("cache_read_input_tokens", 0)
+            it_tok = _usage_int(usage, "input_tokens") + _usage_int(usage, "cache_creation_input_tokens") + _usage_int(usage, "cache_read_input_tokens")
             if it_tok > 0:
                 try: mspan.set_input_tokens(it_tok)
                 except Exception: pass
-            if usage.get("output_tokens", 0) > 0:
-                try: mspan.set_output_tokens(usage["output_tokens"])
+            if _usage_int(usage, "output_tokens") > 0:
+                try: mspan.set_output_tokens(_usage_int(usage, "output_tokens"))
                 except Exception: pass
             mspan_ctx = client.get_span_from_header(mspan.to_header())
             mspan.finish()

package/scripts/codex/cozeloop_hook.py

@@ -787,6 +787,18 @@ def truncate_text(text: str, limit: int = 12000) -> str:
 
 # --- Message Grouping ---
 
+def _usage_int(usage, key):
+    """从 token_usage dict 读 token 数，missing/None/非数字 一律按 0。
+
+    transcript 里 token_usage 字段可能【存在但值为 null】，dict.get(key, 0) 对显式 null
+    返回 None 而非 0，后续 None > 0 / None + int 会抛 TypeError 中断上报。这里统一兜底。
+    """
+    if not isinstance(usage, dict):
+        return 0
+    v = usage.get(key)
+    return v if isinstance(v, int) else 0
+
+
 def _parse_ts(obj):
     """从 codex entry/payload 的 timestamp 解析 datetime（带时区）。失败返回 None。
 
@@ -1299,8 +1311,8 @@ def send_turns_to_cozeloop(turns: List[Dict[str, Any]], session_id: str, model_n
 
                                 # Set token usage
                                 token_usage = turn.get("token_usage", {})
-                                input_tokens = token_usage.get("input_tokens", 0)
-                                output_tokens = token_usage.get("output_tokens", 0)
+                                input_tokens = _usage_int(token_usage, "input_tokens")
+                                output_tokens = _usage_int(token_usage, "output_tokens")
                                 if input_tokens > 0:
                                     model_span.set_input_tokens(input_tokens)
                                 if output_tokens > 0:
@@ -1446,10 +1458,12 @@ def send_turns_to_cozeloop(turns: List[Dict[str, Any]], session_id: str, model_n
                                                     sa_model_span.set_output(ModelOutput(choices=sa_choices))
 
                                                     sa_token = sa_turn.get("token_usage", {})
-                                                    if sa_token.get("input_tokens", 0) > 0:
-                                                        sa_model_span.set_input_tokens(sa_token["input_tokens"])
-                                                    if sa_token.get("output_tokens", 0) > 0:
-                                                        sa_model_span.set_output_tokens(sa_token["output_tokens"])
+                                                    sa_in = _usage_int(sa_token, "input_tokens")
+                                                    sa_out = _usage_int(sa_token, "output_tokens")
+                                                    if sa_in > 0:
+                                                        sa_model_span.set_input_tokens(sa_in)
+                                                    if sa_out > 0:
+                                                        sa_model_span.set_output_tokens(sa_out)
 
                                             # Subagent tool spans
                                             for sa_tc in sa_turn.get("tool_calls", []):