[Autogen Studio UI] Agent not deterministic when using skill

[SrijitaRoy913]

Hi everyone,

I'm using Autogen for a use-case of fetching relevant information from VectorDB. I've created an agent vectordb_assistant that uses skill fetch_relevant_ddl. When this agent is used in a workflow, it does not always invoke the skill correctly.

`from langchain_postgres.vectorstores import PGVector
import datetime
from langchain_community.vectorstores.pgvector import DistanceStrategy
from langchain_community.embeddings import HuggingFaceEmbeddings
from pprint import pprint

doc: langchain_core.documents.base.Document

from langchain_core.documents.base import Document as LangchainDoc
from dataclasses import dataclass

def fetch_relevant_ddl(question: str):
@DataClass
class Document:
page_content: str
metadata: dict

PGVECTOR_CONN_STRING_LOCATE="postgresql+psycopg://username:pswd@host:port/db"

    
schema_vector_db = PGVector.from_existing_index(
    collection_name=collection_name,
    embedding=HuggingFaceEmbeddings(model_name=model_name),
    connection=PGVECTOR_CONN_STRING_LOCATE,
    distance_strategy=DistanceStrategy.COSINE,
    use_jsonb=True)

relevant_tables = schema_vector_db.similarity_search(question, k=5)

relevant_table_list = []

for item in relevant_tables:
    formatted_content = (
                item.page_content.replace("\\n", "\n") +
                item.metadata['table_info'].replace("\\n", "\n").replace("\\t", "\t")
            )
    relevant_table_list.append(formatted_content)     

return relevant_table_list

#Example usage:

from skills import fetch_relevant_ddl

question="How many routes are there?"

results=fetch_relevant_ddl(question)

print(results)`

I've added example usage for it to understand, but it doesn't always invoke the skill correctly on first try.
Any suggestions?