chore(agent-v2): sync changes (#38442)

Co-authored-by: Joel <iamjoel007@gmail.com>
Co-authored-by: zyssyz123 <916125788@qq.com>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Co-authored-by: 林玮 (Jade Lin) <linw1995@icloud.com>
Co-authored-by: 盐粒 Yanli <mail@yanli.one>

api/services/agent/composer_service.py

@@ -33,6 +33,7 @@ from models.workflow import Workflow
 from services.agent.agent_soul_state import agent_soul_has_model
 from services.agent.composer_validator import ComposerConfigValidator
 from services.agent.errors import (
+    AgentModelNotConfiguredError,
     AgentNameConflictError,
     AgentNotFoundError,
     AgentVersionConflictError,
@@ -168,7 +169,8 @@ class AgentComposerService:
 
         _backfill_cli_tool_ids(payload.agent_soul)
         _validate_composer_payload_for_strategy(payload)
-        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
+        if payload.save_strategy in _PUBLISH_SAVE_STRATEGIES:
+            cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
         workflow = cls._get_draft_workflow(tenant_id=tenant_id, app_id=app_id)
         binding = cls._get_workflow_binding(tenant_id=tenant_id, workflow_id=workflow.id, node_id=node_id)
 
@@ -357,7 +359,6 @@ class AgentComposerService:
             raise ValueError("agent_soul is required")
         _backfill_cli_tool_ids(payload.agent_soul)
         _validate_composer_payload_for_strategy(payload)
-        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
 
         agent = cls._get_agent_app_agent(tenant_id=tenant_id, app_id=app_id)
         if not agent:
@@ -401,7 +402,6 @@ class AgentComposerService:
             raise ValueError("agent_soul is required")
         _backfill_cli_tool_ids(payload.agent_soul)
         _validate_composer_payload_for_strategy(payload)
-        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
         agent = cls._require_agent(tenant_id=tenant_id, agent_id=agent_id)
         return cls._save_agent_composer_for_agent(
             tenant_id=tenant_id,
@@ -511,6 +511,8 @@ class AgentComposerService:
                 version_note=version_note,
             )
         )
+        if not agent_soul_has_model(agent_soul):
+            raise AgentModelNotConfiguredError()
         cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=agent_soul)
         version = cls._create_config_version(
             tenant_id=tenant_id,
@@ -591,7 +593,6 @@ class AgentComposerService:
             raise ValueError("agent_soul is required")
         _backfill_cli_tool_ids(payload.agent_soul)
         ComposerConfigValidator.validate_draft_save_payload(payload)
-        cls.validate_knowledge_datasets(tenant_id=tenant_id, agent_soul=payload.agent_soul)
         agent = cls._require_agent(tenant_id=tenant_id, agent_id=agent_id)
         build_draft = cls._save_agent_draft(
             tenant_id=tenant_id,

api/services/agent/composer_validator.py

@@ -3,6 +3,7 @@ from typing import Any
 
 from pydantic import ValidationError
 
+from models.agent_config_entities import AgentKnowledgeQueryMode
 from services.agent.errors import AgentSoulLockedError, InvalidComposerConfigError, PlaintextSecretNotAllowedError
 from services.agent.prompt_mentions import (
     MAX_MENTIONS_PER_PROMPT,
@@ -228,9 +229,40 @@ class ComposerConfigValidator:
     @classmethod
     def validate_agent_soul(cls, agent_soul: AgentSoulConfig) -> None:
         dumped = agent_soul.model_dump(mode="json")
+        cls._validate_knowledge_runtime_config(agent_soul)
         cls._reject_plaintext_secrets(dumped, path="agent_soul")
         cls._validate_shell_config(dumped)
 
+    @classmethod
+    def _validate_knowledge_runtime_config(cls, agent_soul: AgentSoulConfig) -> None:
+        """Validate knowledge settings that are required only for publish/run.
+
+        Draft composer saves must be able to persist partially configured
+        knowledge sets while a user is still editing the panel. These checks
+        stay in the publish validator so invalid runtime configs are still
+        blocked before a version can be published or executed.
+        """
+        for knowledge_set in agent_soul.knowledge.sets:
+            if (
+                knowledge_set.query.mode == AgentKnowledgeQueryMode.USER_QUERY
+                and not (knowledge_set.query.value or "").strip()
+            ):
+                raise InvalidComposerConfigError("knowledge query.value is required for user_query mode")
+
+            retrieval = knowledge_set.retrieval
+            if retrieval.mode == "multiple" and retrieval.top_k is None:
+                raise InvalidComposerConfigError("knowledge retrieval.top_k is required for multiple mode")
+            if retrieval.mode == "single" and retrieval.model is None:
+                raise InvalidComposerConfigError("knowledge retrieval.model is required for single mode")
+
+            metadata_filtering = knowledge_set.metadata_filtering
+            if metadata_filtering.mode == "automatic" and metadata_filtering.metadata_model_config is None:
+                raise InvalidComposerConfigError("metadata_filtering.model_config is required for automatic mode")
+            if metadata_filtering.mode == "manual" and (
+                metadata_filtering.conditions is None or not metadata_filtering.conditions.conditions
+            ):
+                raise InvalidComposerConfigError("metadata_filtering.conditions is required for manual mode")
+
     @classmethod
     def validate_node_job(cls, node_job: WorkflowNodeJobConfig) -> None:
         cls._reject_plaintext_secrets(node_job.model_dump(mode="json"), path="node_job")

api/services/agent/errors.py

@@ -1,5 +1,7 @@
 from werkzeug.exceptions import BadRequest, Conflict, NotFound
 
+from libs.exception import BaseHTTPException
+
 
 class AgentNotFoundError(NotFound):
     description = "Agent not found."
@@ -21,6 +23,12 @@ class AgentVersionConflictError(Conflict):
     description = "Agent config version changed. Please reload and try again."
 
 
+class AgentModelNotConfiguredError(BaseHTTPException):
+    error_code = "agent_model_not_configured"
+    description = "Agent App requires the Agent Soul model to be configured."
+    code = 400
+
+
 class AgentSoulLockedError(BadRequest):
     description = "Agent Soul is locked for this workflow node."
 

api/services/agent_app_sandbox_service.py

@@ -3,12 +3,16 @@
 These services keep product-facing locators (conversation, workflow run, node)
 on the API boundary and translate them into the agent backend's
 ``SandboxLocator`` using persisted non-sensitive runtime layer specs plus the
-saved Agenton session snapshot.
+saved Agenton session snapshot. Upload responses stay console-facing here: the
+agent backend still returns a canonical ToolFile mapping, while this API layer
+re-resolves that mapping into a signed browser download URL.
 """
 
 from __future__ import annotations
 
+import urllib.parse
 from collections.abc import Callable
+from typing import Any
 
 from agenton.compositor import CompositorSessionSnapshot
 from dify_agent.client import Client
@@ -18,7 +22,10 @@ from sqlalchemy import select
 
 from configs import dify_config
 from core.app.apps.agent_app.session_store import AgentAppRuntimeSessionStore
+from core.app.file_access import DatabaseFileAccessController
+from core.app.workflow.file_runtime import DifyWorkflowFileRuntime
 from core.db.session_factory import session_factory
+from factories import file_factory
 from models.agent import AgentRuntimeSessionOwnerType, WorkflowAgentRuntimeSession, WorkflowAgentRuntimeSessionStatus
 
 _RUNTIME_LAYER_SPECS_ADAPTER: TypeAdapter[list[RuntimeLayerSpec]] = TypeAdapter(list[RuntimeLayerSpec])
@@ -45,6 +52,12 @@ class AgentSandboxInfo(BaseModel):
     workspace_cwd: str
 
 
+class AgentSandboxUploadDownload(BaseModel):
+    """Signed browser download URL for one sandbox upload result."""
+
+    url: str
+
+
 class AgentAppSandboxService:
     """Inspect and proxy file access for an Agent App conversation sandbox."""
 
@@ -77,9 +90,15 @@ class AgentAppSandboxService:
         locator = self._resolve_locator(tenant_id=tenant_id, app_id=app_id, conversation_id=conversation_id)
         return self._client_factory().read_sandbox_file_sync(locator, path)
 
-    def upload_file(self, *, tenant_id: str, app_id: str, conversation_id: str, path: str):
+    def upload_file(
+        self, *, tenant_id: str, app_id: str, conversation_id: str, path: str
+    ) -> AgentSandboxUploadDownload:
         locator = self._resolve_locator(tenant_id=tenant_id, app_id=app_id, conversation_id=conversation_id)
-        return self._client_factory().upload_sandbox_file_sync(locator, path)
+        uploaded = self._client_factory().upload_sandbox_file_sync(locator, path)
+        return _upload_download_response(
+            tenant_id=tenant_id,
+            file_mapping=uploaded.file.model_dump(mode="python"),
+        )
 
     def _resolve_locator(self, *, tenant_id: str, app_id: str, conversation_id: str) -> SandboxLocator:
         stored = self._session_store.load_active_session_for_conversation(
@@ -153,7 +172,7 @@ class WorkflowAgentSandboxService:
         node_id: str,
         node_execution_id: str | None,
         path: str,
-    ):
+    ) -> AgentSandboxUploadDownload:
         locator = self._resolve_locator(
             tenant_id=tenant_id,
             app_id=app_id,
@@ -161,7 +180,11 @@ class WorkflowAgentSandboxService:
             node_id=node_id,
             node_execution_id=node_execution_id,
         )
-        return self._client_factory().upload_sandbox_file_sync(locator, path)
+        uploaded = self._client_factory().upload_sandbox_file_sync(locator, path)
+        return _upload_download_response(
+            tenant_id=tenant_id,
+            file_mapping=uploaded.file.model_dump(mode="python"),
+        )
 
     def _resolve_locator(
         self,
@@ -246,6 +269,41 @@ def _deserialize_runtime_layer_specs(value: str | None) -> list[RuntimeLayerSpec
     return _RUNTIME_LAYER_SPECS_ADAPTER.validate_json(value)
 
 
+def _upload_download_response(*, tenant_id: str, file_mapping: dict[str, Any]) -> AgentSandboxUploadDownload:
+    """Resolve one uploaded ToolFile mapping into a signed external download URL."""
+
+    controller = DatabaseFileAccessController()
+    runtime = DifyWorkflowFileRuntime(file_access_controller=controller)
+    try:
+        file = file_factory.build_from_mapping(
+            mapping=file_mapping,
+            tenant_id=tenant_id,
+            access_controller=controller,
+        )
+        url = runtime.resolve_file_url(file=file, for_external=True)
+    except ValueError as exc:
+        raise AgentSandboxInspectorError(
+            "sandbox_upload_download_unavailable",
+            "uploaded sandbox file could not be converted to a download URL",
+            status_code=502,
+        ) from exc
+
+    if not url:
+        raise AgentSandboxInspectorError(
+            "sandbox_upload_download_unavailable",
+            "uploaded sandbox file does not support download URL generation",
+            status_code=502,
+        )
+    return AgentSandboxUploadDownload(url=_with_as_attachment(url))
+
+
+def _with_as_attachment(url: str) -> str:
+    parsed = urllib.parse.urlsplit(url)
+    query = urllib.parse.parse_qsl(parsed.query, keep_blank_values=True)
+    query.append(("as_attachment", "true"))
+    return urllib.parse.urlunsplit(parsed._replace(query=urllib.parse.urlencode(query)))
+
+
 def _default_client_factory() -> Client:
     base_url = dify_config.AGENT_BACKEND_BASE_URL
     if not base_url:
@@ -257,4 +315,10 @@ def _default_client_factory() -> Client:
     return Client(base_url=base_url)
 
 
-__all__ = ["AgentAppSandboxService", "AgentSandboxInfo", "AgentSandboxInspectorError", "WorkflowAgentSandboxService"]
+__all__ = [
+    "AgentAppSandboxService",
+    "AgentSandboxInfo",
+    "AgentSandboxInspectorError",
+    "AgentSandboxUploadDownload",
+    "WorkflowAgentSandboxService",
+]