Is crewAI still unable to complete more complex blogs by executing tasks in order?

[yansircc]

I recently wrote a demo to try and accomplish my blogging mission.
But it's clear that the proxies don't seem to be able to work well with each other.
In particular, I wrote a script to determine if the current text contains all the keywords, but every time it prompts the following error, indicating that two parameters could not be passed to the script:
pydantic.v1.error_wrappers.ValidationError: 1 validation error for Keyword CheckerSchemaSchema checkerSchemaSchema field required (type=value_error.missing)
Here is my demo code:

from crewai import Agent, Task, Crew, Process
from langchain.agents import Tool
from langchain_community.tools import google_serper, tool
from serpapi.google_search import GoogleSearch
from langchain_openai import ChatOpenAI as OpenAI
from langchain_community.llms import ollama
import os

blog_title = "What is BPA free plastic?"
target_word_count = 1000
max_leading_paragraph_word_count = 80
max_snippet_answer_word_count = 50
max_summary_word_count = 50
target_audience = "b2b water bottle buyers"
os.environ["OPENAI_API_KEY"] = "xxx"
# ollama_llm = ollama(model="mistral")

# Define the custom tools
def retrieve_serp_data(query: str) -> dict:
    """
    Returns the serp data for the given query.
    """
    params = {
        "api_key": "xxx",
        "engine": "google",
        "q": query,
        "location": "Austin, Texas, United States",
        "google_domain": "google.com",
        "gl": "us",
        "hl": "en"
    }

    search = GoogleSearch(params)
    results = search.get_dict()

    # Extracting the titles from the organic results
    titles = [item['title'] for item in results.get('organic_results', [])]

    # Extracting questions and snippets from the People Also Ask section
    people_also_ask = [{'question': item.get('question'), 'snippet': item.get('snippet')} 
                       for item in results.get('related_questions', [])]

    # Extracting related keywords
    related_keywords = [item['query'] for item in results.get('related_searches', [])]

    serp_data_json = {
        "titles": titles,
        "people_also_ask": people_also_ask,
        "related_keywords": related_keywords
    }

    return serp_data_json

def word_count(query: str) -> int:
    """
    Returns the word count of the given query.
    """
    # Splitting the query into words based on spaces
    words = query.split()

    # Counting the number of words
    count = len(words)

    return count

def check_keywords(article: str, keywords: list) -> list:
    """
    Checks if all keywords are present in the article.
    Returns a list of missing keywords.
    """
    missing_keywords = [keyword for keyword in keywords if keyword not in article]
    return missing_keywords

# Initialize the tools
# os.environ["SERPER_API_KEY"] = "Your Key"

class CustomTools():
    @tool("Word Count Checker") 
    def word_count(query: str) -> int:
        """
        Returns the word count of the given query.
        """
        # Splitting the query into words based on spaces
        words = query.split()

        # Counting the number of words
        count = len(words)

        return count
    
    @tool("Keyword Checker")
    def check_keywords(article: str, keywords: list) -> list:
        """
        Checks if all keywords are present in the article.
        Returns a list of missing keywords.
        This function accepts 2 must-have parameters: article and keywords.
        article: The text you want to check.
        keywords: A list of keywords you want to check.
        """
        missing_keywords = [keyword for keyword in keywords if keyword not in article]
        return missing_keywords

# Loading Human Tools
# human_tools = load_tools(["human"])

# Retrieve SERP data globally
global_serp_data = retrieve_serp_data(blog_title)
titles = global_serp_data['titles']
people_also_ask = global_serp_data['people_also_ask']
related_keywords = global_serp_data['related_keywords']

# Define the SEO Expert Agent
seo_expert_agent = Agent(
    role='SEO Expert',
    goal='Ensure all related keywords are included in the article.',
    backstory="""You are an SEO expert. Your job is to ensure that the article ranks well on Google.""",
    tools=[CustomTools.check_keywords],
    llm=OpenAI(model_name="gpt-4", temperature=0),
    verbose=True,
    allow_delegation=True
)

# Define the agents
headline_agent = Agent(
    role='Headline Creator',
    goal='Create an engaging, question-based headline.',
    backstory="""You are a headline creator. Your job is to create an engaging, question-based headline.""",
    llm=OpenAI(model_name="gpt-4", temperature=0.7),
    verbose=True,
    allow_delegation=True
)

pas_formula_agent = Agent(
    role='PAS Formula Writer',
    goal='Write the leading paragraph using the PAS formula within 80 words.',
    backstory="""You are a PAS formula writer. Your job is to write the leading paragraph using the PAS formula.""",
    llm=OpenAI(model_name="gpt-4", temperature=0.7),
    verbose=True,
    tools=[CustomTools.word_count],
    allow_delegation=True
)

snippet_answer_agent = Agent(
    role='Snippet Answer Writer',
    goal='Compose the second paragraph directly answering the headline question within 50 words.',
    backstory="""You are a snippet answer writer. Your job is to write a very helpful answer to the headline question for get featured in the snippet box.""",
    llm=OpenAI(model_name="gpt-4", temperature=0.2),
    verbose=True,
    tools=[CustomTools.word_count],
    allow_delegation=True
)

related_questions_agent = Agent(
    role='Related Questions Elaborator',
    goal='List and elaborate on related questions in a structured format.',
    backstory="""You are a related questions elaborator.""",
    llm=OpenAI(model_name="gpt-4", temperature=0.7),
    verbose=True,
    allow_delegation=True
)

summary_agent = Agent(
    role='Summary Writer',
    goal='Conclude the article with a succinct and engaging summary under 50 words.',
    backstory="""You are a summary writer. Your job is to write a succinct and engaging summary.""",
    llm=OpenAI(model_name="gpt-4", temperature=0.7),
    verbose=True,
    tools=[CustomTools.word_count],
    allow_delegation=True
)

editor_agent = Agent(
    role='Editor',
    goal='Ensure the quality and coherence of the entire article.',
    backstory="""You are an experienced editor who will check if the quality of the post meets the requirements and will provide constructive suggestions.""",
    verbose=True,
    tools=[CustomTools.word_count],
    allow_delegation=True
)

# Define the tasks
headline_task = Task(description=f'Create a question-based headline for the article. Try to imitate the following titles of the top ten results on Google SERP as much as possible for better ranking: {titles}. Don\'t forget the target audience are {target_audience}', agent=headline_agent)
pas_formula_task = Task(
    description=f'Write a leading paragraph using the PAS formula for these target audience: {target_audience}. Pick some keywords and insert to your content: {related_keywords}. Check if the text exceeds {max_leading_paragraph_word_count} words, if so, rewrite.',
    agent=pas_formula_agent,
    tools=[CustomTools.word_count]
    )
snippet_answer_task = Task(
    description=f'Write a second paragraph that directly answers the headline question for the target audience: {target_audience}. Pick some keywords and insert to your content: {related_keywords}. Don\'t forget to add a transitional paragraph at the end to keep the readers engaged and encourage them to continue reading. Check if the text exceeds {max_snippet_answer_word_count} words, if so, rewrite.',
    agent=snippet_answer_agent,
    tools=[CustomTools.word_count]
    )
related_questions_task = Task(
    description=f'''List and answer relevant questions for the target audience({target_audience}), enabling them to gain deeper insights into the topic of the current article. Note that pick some keywords and insert to your content: {related_keywords}.
    To achieve better rankings, make sure that all these questions are obtained: {people_also_ask}(The snippets in this list represent the standard answers to the corresponding questions. They should be referred to during the creation process to avoid factual errors).
    To ensure that each paragraph is written long enough, use the PAS formula to make the content more abundant.''',
    agent=related_questions_agent
    )
summary_task = Task(
    description=f'Write a concise and powerful summary. Pick some keywords and insert to your content: {related_keywords}. Check if the text exceeds {max_summary_word_count} words, if so, rewrite.',
    agent=summary_agent,
    tools=[CustomTools.word_count]
    )
editor_task = Task(
    description=f'''Review the article for quality and coherence.
    Check if the post's leading paragraph uses the PAS formula and is within 80 words, if not, ask relevant co-workers to rewrite this paragraph;
	Check if the second paragraph of the post is truly helpful to the readers and within 50 words, if not, ask relevant co-workers to rewrite this paragraph;
	Check if the summary paragraph of the post is within 50 words, if not, ask relevant co-workers to rewrite this paragraph;
	Check if all the article's secondary headings and corresponding content are relevant to the article's theme;
	Check if the total word count of the article  is reaching {target_word_count} words, and if not, ask relevant co-workers to add more secondary headings and corresponding content, while providing your expansion suggestions.''',
    agent=editor_agent,
    tools=[CustomTools.word_count]
    )
seo_expert_task = Task(
    description=f'Check if all related keywords({related_keywords}) appear in the article. If not, try to insert the missing keywords to the article.',
    agent=seo_expert_agent,
    tools=[CustomTools.check_keywords]
    )

# Instantiate the crew with a sequential process
crew = Crew(
    agents=[headline_agent, pas_formula_agent, snippet_answer_agent, related_questions_agent, summary_agent, editor_agent, seo_expert_agent],
    tasks=[headline_task, pas_formula_task, snippet_answer_task, related_questions_task, summary_task, seo_expert_task, editor_task],
    process=Process.sequential,
    verbose=2
)

# Kick off the crew
result = crew.kickoff()

Thanks in advance for this error checking.

[svickers]

See #135 and #110, it is due to the pydantic version currently referenced but there are workarounds.

[joaomdmoura]

This one is new to me, but flagging it as a bug so I can look at it