dify/api/controllers/console/app/skills.py
Yansong Zhang 0c7e7e0c4e feat(api): port Sandbox + VirtualEnvironment + Skill system from feat/support-agent-sandbox (Phase 5-6)
Port the complete infrastructure for agent sandbox execution and skill system:

Sandbox & Virtual Environment (core/sandbox/, core/virtual_environment/):
- Sandbox entity with lifecycle management (ready/failed/cancelled states)
- SandboxBuilder with fluent API for configuring providers
- 5 VM providers: Local, SSH, Docker, E2B, AWS CodeInterpreter
- VirtualEnvironment base with command execution, file transfer, transport layers
- Channel transport: pipe, queue, socket implementations
- Bash session management and DifyCli binary integration
- Storage: archive storage, file storage, noop storage, presign storage
- Initializers: DifyCli, AppAssets, DraftAppAssets, Skills
- Inspector: file browser, archive/runtime source, script utils
- Security: encryption utils, debug helpers

Skill & App Assets (core/skill/, core/app_assets/, core/app_bundle/):
- Skill entity and manager
- App asset accessor, builder pipeline (file, skill builders)
- App bundle source zip extractor
- Storage and converter utilities

API Endpoints:
- CLI API blueprint (controllers/cli_api/) for sandbox callback
- Sandbox provider management (workspace/sandbox_providers)
- Sandbox file browser (console/sandbox_files)
- App asset management (console/app/app_asset)
- Skill management (console/app/skills)
- Storage file endpoints (controllers/files/storage_files)

Services:
- Sandbox service, provider service, file service
- App asset service, app bundle service

Config:
- CliApiConfig, CreatorsPlatformConfig, CollaborationConfig
- FILES_API_URL for sandbox file access

Note: Controller route registration temporarily commented out (marked TODO)
pending resolution of deep dependency chains (socketio, workflow_comment,
command node, etc.). Core sandbox modules are fully ported and syntax-validated.
110 files changed, 10,549 insertions.

Made-with: Cursor
2026-04-08 17:39:02 +08:00

39 lines
1.4 KiB
Python

from flask import request
from flask_restx import Resource
from controllers.console import console_ns
from controllers.console.app.wraps import get_app_model
from controllers.console.wraps import account_initialization_required, current_account_with_tenant, setup_required
from libs.login import login_required
from models import App
from models.model import AppMode
from services.skill_service import SkillService
@console_ns.route("/apps/<uuid:app_id>/workflows/draft/nodes/llm/skills")
class NodeSkillsApi(Resource):
"""Extract tool dependencies from an LLM node's skill prompts.
The client sends the full node ``data`` object in the request body.
The server real-time builds a ``SkillBundle`` from the current draft
``.md`` assets and resolves transitive tool dependencies — no cached
bundle is used.
"""
@setup_required
@login_required
@account_initialization_required
@get_app_model(mode=[AppMode.ADVANCED_CHAT, AppMode.WORKFLOW])
def post(self, app_model: App):
current_user, _ = current_account_with_tenant()
node_data = request.get_json(force=True)
if not isinstance(node_data, dict):
return {"tool_dependencies": []}
tool_deps = SkillService.extract_tool_dependencies(
app=app_model,
node_data=node_data,
user_id=current_user.id,
)
return {"tool_dependencies": [d.model_dump() for d in tool_deps]}