feat(telemetry): add isolated beta quality diagnostics and release identity

src/langbot/pkg/plugin/agent_run_support.py

@@ -2,6 +2,8 @@
 
 from __future__ import annotations
 
+from ..telemetry import diagnostics
+
 from typing import Any, Union
 import json
 import time
@@ -302,6 +304,7 @@ async def _validate_agent_run_session(
         if not available_apis.get(api_capability, False) and not has_admin_permission:
             return None, handler.ActionResponse.error(message=f'{api_name} access not authorized')
 
+    diagnostics.link_context(session.get('_diagnostic_context'))
     return session, None
 
 

src/langbot/pkg/plugin/connector.py

@@ -1,6 +1,8 @@
 # For connect to plugin runtime.
 from __future__ import annotations
 
+from langbot.pkg.telemetry import diagnostics
+
 import asyncio
 import contextlib
 import contextvars
@@ -668,6 +670,7 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
                         }
                     )
 
+    @diagnostics.observe('lifecycle', 'runtime.prepare_connected_runtime', source='runtime', stage='execute')
     async def _prepare_connected_runtime(self) -> None:
         """Handshake follow-up: pin OSS compatibility, then replay authority."""
 
@@ -853,6 +856,7 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
                     self.schedule_reconnect()
                     failures = 0
 
+    @diagnostics.observe('lifecycle', 'runtime.initialize', source='runtime', stage='execute')
     async def initialize(self):
         if not self.is_enable_plugin:
             self.ap.logger.info('Plugin system is disabled.')
@@ -1705,6 +1709,7 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
             )
             return plugin_package, latest_version
 
+    @diagnostics.observe('lifecycle', 'runtime.install_plugin', source='runtime', stage='execute')
     async def install_plugin(
         self,
         install_source: PluginInstallSource,
@@ -1820,6 +1825,7 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
             task_context.set_current_action('plugin updated' if operation == 'upgrade' else 'plugin installed')
             task_context.metadata['progress_percent'] = 100
 
+    @diagnostics.observe('lifecycle', 'runtime.upgrade_plugin', source='runtime', stage='execute')
     async def upgrade_plugin(
         self,
         plugin_author: str,
@@ -1846,6 +1852,7 @@ class PluginRuntimeConnector(ManagedRuntimeConnector):
         )
         return {}
 
+    @diagnostics.observe('lifecycle', 'runtime.delete_plugin', source='runtime', stage='execute')
     async def delete_plugin(
         self,
         plugin_author: str,

src/langbot/pkg/plugin/handler.py

@@ -1,5 +1,7 @@
 from __future__ import annotations
 
+from ..telemetry import diagnostics
+
 import asyncio
 import inspect
 import typing
@@ -639,6 +641,7 @@ class RuntimeConnectionHandler(handler.Handler):
         avoids reserving one pooled connection across provider and network waits.
         """
 
+        diagnostics.annotate(workspace_uuid=action_context.workspace_uuid)
         persistence_mgr = getattr(self.ap, 'persistence_mgr', None)
         if persistence_mgr is None:
             yield
@@ -687,10 +690,13 @@ class RuntimeConnectionHandler(handler.Handler):
                             return
                         if trusted_plugin_identity is not None:
                             safe_data['caller_plugin_identity'] = trusted_plugin_identity
-                        async for response in _action_handler(safe_data):
-                            yield response
+                        async with contextlib.aclosing(_action_handler(safe_data)) as responses:
+                            async for response in responses:
+                                yield response
 
-                self.actions[action_name] = secured_stream_action
+                self.actions[action_name] = diagnostics.observe(
+                    'api', 'host.' + action_name, source='plugin', ap=self.ap
+                )(secured_stream_action)
                 continue
 
             async def secured_action(
@@ -719,7 +725,9 @@ class RuntimeConnectionHandler(handler.Handler):
                         response = await response
                     return response
 
-            self.actions[action_name] = secured_action
+            self.actions[action_name] = diagnostics.observe('api', 'host.' + action_name, source='plugin', ap=self.ap)(
+                secured_action
+            )
 
     async def _get_plugin_setting(
         self,