task generation (#450)

Co-authored-by: Shuchang Zheng <wintonzheng0325@gmail.com>
This commit is contained in:
Kerem Yilmaz
2024-06-07 15:59:53 -07:00
committed by GitHub
parent 12b83e009e
commit d18fc5b59c
8 changed files with 226 additions and 4 deletions

View File

@@ -0,0 +1,16 @@
We are building an AI agent that can automate browser tasks. The task creation schema is a JSON object with the following fields:
url: required field, the starting URL for the task. This will be the first page the agent visits in order to achieve its goals.
navigation_goal: optional. The value should be a string that we can use as an input to a Large Language Modal. It needs to tell the agent the goal in terms of navigating the website. It needs to define a single goal. You can include explicit completion and failure criteria. You can define guardrails that could help the agent from taking certain actions or getting derailed.
data_extraction_goal: optional. The value should be a string that we can use as an input to a Large Language Modal. It needs to tell the agent the goal in terms of extracting data. It needs to be a single goal.
navigation_payload: optional. The value should be JSON. Use this field if there is any information for the agent to be able to complete the task such as values that can help fill a form, parameters for queries and so on.
extracted_information_schema: optional. The exact schema of the data to be extracted.
At least one of navigation goal or data extraction goal should be provided. The agent can't proceed without any goals.
If a field is not required to achieve a task, provide the value `null`.
Respond with only JSON output that follows the task creation schema for the following prompt:
```
{{ user_prompt }}
```