coze_lab 0.1.47 → 0.1.49

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "coze_lab",
-  "version": "0.1.47",
+  "version": "0.1.49",
   "description": "Configure local AI agents (Claude Code, Codex, OpenClaw) to report traces to CozeLoop",
   "keywords": [
     "cozeloop",

package/scripts/claude-code/cozeloop_hook.py

@@ -1472,7 +1472,12 @@ def send_steps_realtime(turns, session_id, history_turns, state, coze_tags_overr
     if token:
         os.environ["COZELOOP_API_TOKEN"] = token
     workspace_id = os.environ.get("COZELOOP_WORKSPACE_ID", "") or _DEFAULT_WORKSPACE_ID
-    client_kwargs = {"ultra_large_report": True, "upload_timeout": 120}
+    upload_events: List[str] = []
+    client_kwargs = {
+        "ultra_large_report": True,
+        "upload_timeout": 120,
+        "trace_finish_event_processor": _make_finish_event_processor(upload_events),
+    }
     if workspace_id:
         client_kwargs["workspace_id"] = workspace_id
     if token:
@@ -1539,6 +1544,9 @@ def send_steps_realtime(turns, session_id, history_turns, state, coze_tags_overr
             # 实测：root finish 后，子 span 仍可用其 header 挂到同一 trace_id 下，不影响后续增量。
             root_span.finish()
             client.flush()
+            if upload_events:
+                debug_log(f"[rt] upload failed after root flush: {upload_events[-1][:500]}")
+                return None
             debug_log(f"[rt] root created+finished trace_id={getattr(root_ctx,'trace_id','?')}")
 
         # ---- 发新完成的 step ----
@@ -1622,6 +1630,9 @@ def send_steps_realtime(turns, session_id, history_turns, state, coze_tags_overr
             sent += 1
             # 每个 step 立即 flush —— 这是“结束即可见”的关键。
             client.flush()
+            if upload_events:
+                debug_log(f"[rt] upload failed after step flush: {upload_events[-1][:500]}")
+                return None
 
         new_last = last_global + sent
         state["rt_last_global_step"] = new_last
@@ -1650,8 +1661,15 @@ def send_steps_realtime(turns, session_id, history_turns, state, coze_tags_overr
                 fin.set_output(last_output)
                 fin.finish()
             client.flush()
+            if upload_events:
+                debug_log(f"[rt] upload failed after final flush: {upload_events[-1][:500]}")
+                return None
             debug_log(f"[rt] finalized, total sent steps={new_last}")
 
+        if upload_events:
+            debug_log(f"[rt] upload failed, state not advanced. Last failure: {upload_events[-1][:500]}")
+            return None
+
         debug_log(f"[rt] sent {sent} new step(s), last_global={new_last}/{total_steps}, terminal={is_terminal}")
         return True
     except Exception as e:

package/scripts/codex/cozeloop_hook.py

@@ -131,6 +131,127 @@ else:
 
 # --- Configuration ---
 DEBUG = os.environ.get("CC_COZELOOP_DEBUG", "").lower() == "true"
+_DEFAULT_WORKSPACE_ID = "7649231955045072915"  # hardcoded spaceID fallback
+
+# 实时 tool span 上报的批量间隔（秒）：距上次上报不足此值则本次不发，把已结束的 tool span
+# 攒到下次一起发（最多每 5s 一批）。终态(Stop/SubagentStop)不受限，立即收尾。
+REALTIME_BATCH_INTERVAL = float(os.environ.get("COZELOOP_REALTIME_BATCH_INTERVAL", "5"))
+
+
+# --- coze-context parsing -------------------------------------------------
+# User messages may embed a block like:
+#   <coze-context>
+#   account_id: 0
+#   agent_id: 7644920552473395499
+#   session_id: 7644919579054997796
+#   message_id: 04dd5246-...
+#   </coze-context>
+# We parse its key:value pairs and inject them into the trace.
+
+_COZE_CTX_OPEN = "<coze-context>"
+_COZE_CTX_CLOSE = "</coze-context>"
+
+
+def parse_coze_context(text: str) -> Dict[str, str]:
+    """Extract the LAST <coze-context> block's key:value pairs from text.
+
+    Returns {} if no block is present. Tag keys are prefixed with
+    'coze_' by the caller; here we return raw keys as written.
+    """
+    if not text or _COZE_CTX_OPEN not in text:
+        return {}
+    # Take the last occurrence (latest context wins).
+    open_idx = text.rfind(_COZE_CTX_OPEN)
+    close_idx = text.find(_COZE_CTX_CLOSE, open_idx)
+    if close_idx == -1:
+        return {}
+    body = text[open_idx + len(_COZE_CTX_OPEN):close_idx]
+    # The block may arrive with real newlines, OR with literal backslash-n
+    # (e.g. when the whole message is an embedded JSON string that was never
+    # un-escaped). Normalize both forms before splitting into lines.
+    body = body.replace("\\r\\n", "\n").replace("\\n", "\n").replace("\\r", "\n")
+    result: Dict[str, str] = {}
+    for line in body.splitlines():
+        line = line.strip()
+        if not line or ":" not in line:
+            continue
+        key, _, value = line.partition(":")
+        key = key.strip()
+        value = value.strip()
+        if key:
+            result[key] = value
+    return result
+
+
+def turn_coze_context(turn: Dict[str, Any]) -> Dict[str, str]:
+    """Extract coze-context from a grouped turn with fallbacks for Codex rollout shapes."""
+    texts = [turn.get("user_message_text", "")]
+    for msg in turn.get("input_messages", []):
+        if isinstance(msg, dict):
+            content = msg.get("content", "")
+            if isinstance(content, str):
+                texts.append(content)
+            elif isinstance(content, list):
+                texts.append(extract_message_content_text({"content": content}))
+            elif isinstance(content, dict):
+                texts.append(extract_message_content_text({"content": [content]}))
+    user_payload = turn.get("user_message")
+    if isinstance(user_payload, dict):
+        texts.append(extract_message_content_text(user_payload))
+    for text in texts:
+        ctx = parse_coze_context(text)
+        if ctx:
+            return ctx
+    return {}
+
+
+# --- trace upload failure / logid capture ---------------------------------
+def _extract_logid(msg: str) -> str:
+    """Pull the server logid out of an SDK error message, if present.
+
+    SDK failure messages embed it as 'logid=XXXX' (sometimes within brackets).
+    """
+    if not msg:
+        return ""
+    marker = "logid="
+    idx = msg.find(marker)
+    if idx == -1:
+        return ""
+    rest = msg[idx + len(marker):]
+    logid = []
+    for ch in rest:
+        if ch.isalnum():
+            logid.append(ch)
+        else:
+            break
+    return "".join(logid)
+
+
+def _make_finish_event_processor(upload_events: Optional[List[str]] = None):
+    """Return a trace_finish_event_processor that surfaces failures + logid.
+
+    The CozeLoop SDK calls this for each flush event; on failure we print the
+    server logid to stderr so it can be handed to platform support for tracing
+    the root cause (e.g. via `bytedcli log get-logid-log <logid>`).
+    """
+    def _processor(info):
+        try:
+            if not getattr(info, "is_event_fail", False):
+                hook_log("upload success")
+                return
+            detail = getattr(info, "detail_msg", "") or ""
+            if upload_events is not None:
+                upload_events.append(detail or "trace export failed")
+            logid = _extract_logid(detail)
+            if logid:
+                hook_log(f"upload failed logid={logid} detail={detail[:500]}")
+                print(f"[CozeLoop] 上报失败 logid={logid} (可用 bytedcli log get-logid-log {logid} 排查)", file=sys.stderr)
+            else:
+                hook_log(f"upload failed detail={detail[:500]}")
+                print(f"[CozeLoop] 上报失败: {detail[:300]}", file=sys.stderr)
+        except Exception:
+            pass
+    return _processor
 
 
 def _normalize_api_base_url(raw: str) -> str:
@@ -1251,7 +1372,12 @@ def send_steps_realtime(turns, session_id, state, model_name="codex", coze_tags_
     if token:
         os.environ["COZELOOP_API_TOKEN"] = token
     workspace_id = os.environ.get("COZELOOP_WORKSPACE_ID", "") or _DEFAULT_WORKSPACE_ID
-    client_kwargs = {"ultra_large_report": True, "upload_timeout": 120}
+    upload_events: List[str] = []
+    client_kwargs = {
+        "ultra_large_report": True,
+        "upload_timeout": 120,
+        "trace_finish_event_processor": _make_finish_event_processor(upload_events),
+    }
     if workspace_id:
         client_kwargs["workspace_id"] = workspace_id
     if token:
@@ -1307,6 +1433,9 @@ def send_steps_realtime(turns, session_id, state, model_name="codex", coze_tags_
             root_ctx = client.get_span_from_header(state["rt_root_header"])
             root_span.finish()  # 立即 finish 落库（后端按 root_span 查），子 span 仍可挂同 trace
             client.flush()
+            if upload_events:
+                hook_log(f"[rt] upload failed after root flush detail={upload_events[-1][:500]}")
+                return None
             hook_log(f"[rt] root created trace_id={getattr(root_ctx,'trace_id','?')}")
 
         # 线性展开所有 tool_call（带 result 的才算完成），按全局序增量发
@@ -1361,6 +1490,9 @@ def send_steps_realtime(turns, session_id, state, model_name="codex", coze_tags_
             tspan.finish()
             sent += 1
             client.flush()  # 每个 tool_call 立即 flush → 结束即可见
+            if upload_events:
+                hook_log(f"[rt] upload failed after tool flush detail={upload_events[-1][:500]}")
+                return None
 
         state["rt_last_tool"] = last_tool + sent
         if sent > 0:
@@ -1382,8 +1514,15 @@ def send_steps_realtime(turns, session_id, state, model_name="codex", coze_tags_
                 fin.set_output(truncate_text(last_out))
                 fin.finish()
             client.flush()
+            if upload_events:
+                hook_log(f"[rt] upload failed after final flush detail={upload_events[-1][:500]}")
+                return None
             hook_log(f"[rt] finalized total_tools={state['rt_last_tool']}")
 
+        if upload_events:
+            hook_log(f"[rt] upload failed state not advanced detail={upload_events[-1][:500]}")
+            return None
+
         hook_log(f"[rt] sent {sent} tool span(s), last_tool={state['rt_last_tool']}, terminal={is_terminal}")
         return True
     except Exception as e: