fix(api): resolve all remaining known issues

1. Fix workflow-level total_tokens=0:
   Call graph_runtime_state.add_tokens(usage.total_tokens) in both
   _run_without_tools and _run_with_tools paths after node execution.
   Previously only graphon's internal ModelInvokeCompletedEvent handler
   called add_tokens, which agent-v2 doesn't emit.

2. Fix Turn 2 SSE empty response:
   Set PUBSUB_REDIS_CHANNEL_TYPE=streams in .env. Redis Streams
   provides durable event delivery (consumers can replay past events),
   solving the pub/sub at-most-once timing issue.

3. Skill -> Agent runtime integration:
   SandboxBuilder.build() now auto-includes SkillInitializer if not
   already present. This ensures sandbox.attrs has the skill bundle
   loaded for downstream consumers (tool execution in sandbox).

4. LegacyResponseAdapter:
   New module at core/app/apps/common/legacy_response_adapter.py.
   Filters workflow-specific SSE events (workflow_started, node_started,
   node_finished, workflow_finished) from the stream, passing through
   only message/message_end/agent_log/error/ping events that old
   clients expect.

46 unit tests pass.

Made-with: Cursor

api/core/app/apps/common/legacy_response_adapter.py

@@ -0,0 +1,54 @@
+"""Legacy Response Adapter for transparent upgrade.
+
+When old apps (chat/completion/agent-chat) run through the Agent V2
+workflow engine via transparent upgrade, the SSE events are in workflow
+format (workflow_started, node_started, etc.). This adapter filters out
+workflow-specific events and passes through only the events that old
+clients expect (message, message_end, etc.).
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+from collections.abc import Generator
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+WORKFLOW_ONLY_EVENTS = frozenset({
+    "workflow_started",
+    "workflow_finished",
+    "node_started",
+    "node_finished",
+    "iteration_started",
+    "iteration_next",
+    "iteration_completed",
+})
+
+
+def adapt_workflow_stream_for_legacy(
+    stream: Generator[str, None, None],
+) -> Generator[str, None, None]:
+    """Filter workflow-specific SSE events from a streaming response.
+
+    Passes through message, message_end, agent_log, error, ping events.
+    Suppresses workflow_started, workflow_finished, node_started, node_finished.
+
+    This makes the SSE stream look more like what old easy-UI apps produce,
+    while still carrying the actual LLM response content.
+    """
+    for chunk in stream:
+        if not chunk or not chunk.strip():
+            yield chunk
+            continue
+
+        try:
+            if chunk.startswith("data: "):
+                data = json.loads(chunk[6:])
+                event = data.get("event", "")
+                if event in WORKFLOW_ONLY_EVENTS:
+                    continue
+            yield chunk
+        except (json.JSONDecodeError, TypeError):
+            yield chunk

api/core/sandbox/builder.py

@@ -141,7 +141,16 @@ class SandboxBuilder:
                 assets_id=self._assets_id,
             )
 
-            for init in self._initializers:
+            all_initializers = list(self._initializers)
+            try:
+                from core.sandbox.initializer.skill_initializer import SkillInitializer
+
+                if not any(isinstance(i, SkillInitializer) for i in all_initializers):
+                    all_initializers.append(SkillInitializer())
+            except ImportError:
+                pass
+
+            for init in all_initializers:
                 if isinstance(init, SyncSandboxInitializer):
                     init.initialize(sandbox, ctx)
         except Exception as exc:

api/core/workflow/nodes/agent_v2/node.py

@@ -184,6 +184,7 @@ class AgentV2Node(Node[AgentV2NodeData]):
                 metadata[WorkflowNodeExecutionMetadataKey.TOTAL_TOKENS] = usage.total_tokens
                 metadata[WorkflowNodeExecutionMetadataKey.TOTAL_PRICE] = usage.total_price
                 metadata[WorkflowNodeExecutionMetadataKey.CURRENCY] = usage.currency
+                self.graph_runtime_state.add_tokens(usage.total_tokens)
 
             yield StreamCompletedEvent(
                 node_run_result=NodeRunResult(
@@ -262,6 +263,9 @@ class AgentV2Node(Node[AgentV2NodeData]):
                 node_execution_id=self.id,
             )
 
+            if result.usage and hasattr(result.usage, "total_tokens"):
+                self.graph_runtime_state.add_tokens(result.usage.total_tokens)
+
             yield StreamCompletedEvent(
                 node_run_result=NodeRunResult(
                     status=WorkflowNodeExecutionStatus.SUCCEEDED,