add endpoint/logic for creating a taskv2-workflow from a prompt (#3352)

This commit is contained in:
Jonathan Dobson
2025-09-03 16:55:15 -04:00
committed by GitHub
parent c4934153b6
commit 8b163ba371
3 changed files with 124 additions and 0 deletions

View File

@@ -0,0 +1,14 @@
Given the user goal, come up with a title for it, and a block label.
MAKE SURE YOU OUTPUT VALID JSON. No text before or after JSON, no trailing commas, no comments (//), no unnecessary quotes, etc.
Reply in JSON format with the following keys:
{
"title": str, // A descriptive and informative title for the goal. Use no more than 5 words
"block_label": str, // A label for the block. Use 1 word only. Based off of the "title".
}
User goal:
```
{{ user_goal }}
```

View File

@@ -525,6 +525,55 @@ async def create_workflow(
raise FailedToCreateWorkflow(str(e))
@base_router.post(
"/workflows/create-from-prompt",
include_in_schema=False,
)
async def create_workflow_from_prompt(
data: TaskV2Request,
organization: Organization = Depends(org_auth_service.get_current_org),
x_max_iterations_override: Annotated[int | str | None, Header()] = None,
x_max_steps_override: Annotated[int | str | None, Header()] = None,
) -> dict[str, Any]:
if x_max_iterations_override or x_max_steps_override:
LOG.info(
"Overriding max steps for workflow-from-prompt",
max_iterations_override=x_max_iterations_override,
max_steps_override=x_max_steps_override,
)
await PermissionCheckerFactory.get_instance().check(organization, browser_session_id=data.browser_session_id)
if isinstance(x_max_iterations_override, str):
try:
x_max_iterations_override = int(x_max_iterations_override)
except ValueError:
x_max_iterations_override = None
if isinstance(x_max_steps_override, str):
try:
x_max_steps_override = int(x_max_steps_override)
except ValueError:
x_max_steps_override = None
try:
workflow = await app.WORKFLOW_SERVICE.create_workflow_from_prompt(
organization=organization,
user_prompt=data.user_prompt,
totp_identifier=data.totp_identifier,
totp_verification_url=data.totp_verification_url,
webhook_callback_url=data.webhook_callback_url,
proxy_location=data.proxy_location,
max_screenshot_scrolling_times=data.max_screenshot_scrolls,
extra_http_headers=data.extra_http_headers,
max_iterations=x_max_iterations_override,
max_steps=x_max_steps_override,
)
except Exception as e:
LOG.error("Failed to create workflow from prompt", exc_info=True, organization_id=organization.organization_id)
raise FailedToCreateWorkflow(str(e))
return workflow.model_dump(by_alias=True)
@legacy_base_router.put(
"/workflows/{workflow_id}",
openapi_extra={

View File

@@ -1,6 +1,7 @@
import asyncio
import base64
import json
import uuid
from datetime import UTC, datetime
from typing import Any
@@ -25,6 +26,7 @@ from skyvern.exceptions import (
WorkflowRunNotFound,
)
from skyvern.forge import app
from skyvern.forge.prompts import prompt_engine
from skyvern.forge.sdk.artifact.models import ArtifactType
from skyvern.forge.sdk.core import skyvern_context
from skyvern.forge.sdk.core.security import generate_skyvern_webhook_headers
@@ -110,6 +112,9 @@ from skyvern.webeye.browser_factory import BrowserState
LOG = structlog.get_logger()
DEFAULT_FIRST_BLOCK_LABEL = "block_1"
DEFAULT_WORKFLOW_TITLE = "New Workflow"
class WorkflowService:
@staticmethod
@@ -669,6 +674,62 @@ class WorkflowService:
ai_fallback=False if ai_fallback is None else ai_fallback,
)
async def create_workflow_from_prompt(
self,
organization: Organization,
user_prompt: str,
totp_identifier: str | None = None,
totp_verification_url: str | None = None,
webhook_callback_url: str | None = None,
proxy_location: ProxyLocation | None = None,
max_screenshot_scrolling_times: int | None = None,
extra_http_headers: dict[str, str] | None = None,
max_iterations: int | None = None,
max_steps: int | None = None,
) -> Workflow:
metadata_prompt = prompt_engine.load_prompt(
"conversational_ui_goal",
user_goal=user_prompt,
)
metadata_response = await app.LLM_API_HANDLER(
prompt=metadata_prompt,
prompt_name="conversational_ui_goal",
)
block_label: str = metadata_response.get("block_label", DEFAULT_FIRST_BLOCK_LABEL)
title: str = metadata_response.get("title", DEFAULT_WORKFLOW_TITLE)
task_v2_block = TaskV2Block(
prompt=user_prompt,
totp_identifier=totp_identifier,
totp_verification_url=totp_verification_url,
label=block_label,
max_iterations=max_iterations or settings.MAX_ITERATIONS_PER_TASK_V2,
max_steps=max_steps or settings.MAX_STEPS_PER_TASK_V2,
output_parameter=OutputParameter(
output_parameter_id=str(uuid.uuid4()),
key=f"{block_label}_output",
workflow_id="",
created_at=datetime.now(UTC),
modified_at=datetime.now(UTC),
),
)
new_workflow = await self.create_workflow(
title=title,
workflow_definition=WorkflowDefinition(parameters=[], blocks=[task_v2_block]),
organization_id=organization.organization_id,
proxy_location=proxy_location,
webhook_callback_url=webhook_callback_url,
totp_verification_url=totp_verification_url,
totp_identifier=totp_identifier,
max_screenshot_scrolling_times=max_screenshot_scrolling_times,
extra_http_headers=extra_http_headers,
)
return new_workflow
async def get_workflow(self, workflow_id: str, organization_id: str | None = None) -> Workflow:
workflow = await app.DATABASE.get_workflow(workflow_id=workflow_id, organization_id=organization_id)
if not workflow: