agent.py

from typing import TypedDict,Sequence,Annotated
from dotenv import load_dotenv
from langgraph.graph import StateGraph,START,END
from langchain_core.messages import BaseMessage,SystemMessage,HumanMessage
from langchain_core.tools import tool
from langgraph.prebuilt import ToolNode
from langgraph.graph.message import add_messages
from langchain_google_genai import ChatGoogleGenerativeAI
import asyncio
from google import genai
from gemini import extract_topics_from_tweets,extract_from_prompt,create_content_for_readme,decide_with_stars_are_less,generate_post,get_recent_topics
from gemini import extract_repo_name_from_inp
from twitter import retrieve_tweets_by_query,post_tweets
from github_functions import Readme,list_repos,get_stars
from gemini import decide_intermediate_step_using_msg,generate_about_repo
load_dotenv()

client = genai.Client()

class AgentState(TypedDict):
    messages:Annotated[Sequence[BaseMessage],add_messages]

tools = []

graph = StateGraph(AgentState)
llm = ChatGoogleGenerativeAI(model='gemini-2.5-flash')
def agent(state:AgentState):
    instruction = SystemMessage(content="""
        Consider the user's task and based on that:
        0 → Insight Agent - finds trending dev topics, keywords, opportunities.
        1-> About Repo Agent- generates a two-line description of the repo(Only for github).
        2 → Content Agent - Drafts changes for the readme files(Only for github).
        3 → Design Helper Agent - creates diagrams, visuals, infographics.
        4 → Distribution Agent - posts/schedules across linkedIn, Twitter and not Github.
        5 → Feedback Agent - analyzes engagement (stars, likes, comments).

        The order of actions is Feedback->Insight->About Repo->Content->Distribution    

        Donot execute these agents more than once(no repetetion)    

        Also Don't ever move backwards(Using the agents preceeding the order) meaning if you have executed the content agent, don't execute the insight agent again.

        This order is wrong: Insight-> Feedback
        This order is wrong: Content-> About Repo
        
        For example if you need to execute the content agent, insight agent must be already executed with the positive response.
        
        If you feel you have fulfilled the user's question return END(No INDEX NUMBER) ie, your actions fulfill the user's requirements.
        You don't have to perform any action beyond the user's requirements.
        
        Return with the index of the agent selected
            """)
    response = llm.invoke([instruction] + state['messages'])
    with open("data/user_ques.txt",'w') as f:
        f.write(state['messages'][-1].content)
    return {"messages":response}
def decide(state:AgentState):
    if state["messages"][-1].content == '0':

        return "insights"
    elif state["messages"][-1].content == '1':
        return "about-repo"
    elif state["messages"][-1].content == '2':
        return "gen-content"
    elif state["messages"][-1].content == '3':
        return "design"
    elif state["messages"][-1].content == '4':
        return "posts"
    elif state["messages"][-1].content == '5':
        return "feedback"
    else:
        return "END"
graph.add_node("agent",agent)
def gen_content(state:AgentState):
    with open("data/insights.txt") as f:
        topics = f.read()
    
    #topics = eval(topics)
    # We're testing now
    #topics = ["Agentic AI","Usage of LLM's","LangGraph","Autonomous systems"]
    # Create a function in gemini two extract the repo link.
    res = extract_repo_name_from_inp(state['messages'])
    if res == "False":

        repos = list_repos('Anish-CodeDev')
        for repo in repos:

            readme  = Readme(repo)
            content = readme.load_readme()
            new_content = create_content_for_readme(content,topics)
            print("I am suggesting an improvement, press Y if you like it and N if you don't like it")
            print(new_content)
            inp = input("Your choice: ")
            if inp == "Y":

            
                res = readme.update_readme(new_content)

            elif inp == "N":
                res = 'The user did\'nt like it'
                print(res)
        
        if res != 'The user did\'nt like it':
            return {"messages":"The user did'nt like it"}
    else:
            readme  = Readme(res)
            content = readme.load_readme()
            new_content = create_content_for_readme(content,topics)
            print("I am suggesting an improvement, press Y if you like it and N if you don't like it")
            print(new_content)
            inp = input("Your choice: ")
            if inp == "Y":

            
                res = readme.update_readme(new_content)

            elif inp == "N":
                res = 'The user did\'nt like it'
                print(res)
    return {"messages":"The content of the README of the repo has been updated, what else do you want me to do"}
def gen_insights(state:AgentState):
    topic = extract_from_prompt(state['messages'])
    topics = get_recent_topics(topic)
    with open("data/insights.txt",'w') as f:
        for topic in topics:

            f.write(topic + '\n')

    return {"messages":"The insights have been generated."}
def gen_design(state:AgentState):
    return {"messages":"design"}
def posts(state:AgentState):
    with open("data/repos_to_publicise.txt",'r') as f:
        repos_list = f.read()
    res = extract_repo_name_from_inp(state['messages'])
    if res == "False":
        repos_list = repos_list.split('\n')
        repos_list.remove('')
        for repo in repos_list:
            content = generate_post(repo,"X")
            print(content)

        decision = input("Press Y if you want me to publish the post and press N if you did'nt like the post")
        if decision == "Y":

            res = post_tweets(content)
    else:
        content = generate_post(res,"X")
        print(content)

        decision = input("Press Y if you want me to publish the post and press N if you did'nt like the post")
        if decision == "Y":

            res = post_tweets(content)
    return {"messages":"The draft was shown to you and decision was taken based on your will"}
def manage_feedback(state:AgentState):
    starred = get_stars('Anish-CodeDev')
    with open('data/repos_to_publicise.txt','r+') as f:
        f.truncate(0)
    for repo in starred:
        decision = decide_with_stars_are_less(starred[repo])
        if decision == "less":
            with open('data/repos_to_publicise.txt','a') as f:
                f.write(repo + '\n')
        else:
            print('It was good')
    return {"messages":"The repos with less numbers of stars have been noted"}

def about_repo(state:AgentState):
    with open("data/repos_to_publicise.txt",'r') as f:
        repos_list = f.read()
    res = extract_repo_name_from_inp(state['messages'])
    with open("data/insights.txt") as f:
        topics = f.read()
    topics = topics.split('\n')
    if res == "False":
        repos_list = repos_list.split('\n')
        repos_list.remove('')
        for repo in repos_list:
            content = generate_about_repo(repo,topics)
            print(content)

            decision = input("Press Y if you want me to publish the post and press N if you did'nt like the post")
            if decision == "Y":

                readme = Readme(repo)
                res = readme.update_about(content)
            
    else:
        content = generate_about_repo(res,topics)
        print(content)
        decision = input("Press Y if you want me to publish the post and press N if you did'nt like the post")
        if decision == "Y":
            readme = Readme(res)
            res = readme.update_about(content)
            return {"messages":res}
            
    return {"messages":"about repo"}
graph.add_node("gen_insights",gen_insights)
graph.add_node("content",gen_content)
graph.add_node("design",gen_design)
graph.add_node("posts",posts)
graph.add_node("feedback",manage_feedback)
graph.add_node("about_repo",about_repo)
graph.add_edge(START,"agent")
# Later add a return value for communication between agents
graph.add_conditional_edges(
    "agent",
    decide,
    {
        "gen-content":"content",
        "insights":"gen_insights",
        "design":"design",
        "posts":"posts",
        "feedback":"feedback",
        "about-repo":"about_repo",
        "END":END
    }
)
# When things start working also trying to loop the agent communication
graph.add_edge("content","intermediate")
graph.add_edge("gen_insights","intermediate")
graph.add_edge("design","intermediate")
graph.add_edge("posts","intermediate")
graph.add_edge("feedback","intermediate")
graph.add_edge("about_repo","intermediate")
'''
graph.add_edge("gen_insights","content")
graph.add_edge("posts","feedback")
graph.add_edge("feedback",END)'''

def dummy_func_for_intermediate_node(state:AgentState):
    return {"messages":state['messages']}
graph.add_node("intermediate",dummy_func_for_intermediate_node)
def decide_intermediate_step(state:AgentState):
    with open("data/user_ques.txt",'r') as f:
        request = f.read()
    res = decide_intermediate_step_using_msg(request,state['messages'][-1])
    if res == '0':
        return "insights"
    if res == '1':
        return "about-repo"
    if res == '2':
        return "gen-content"
    if res == '4':
        return "posts"
    
    if res == "5":
        return "feedback"
    else:
        return "END"
graph.add_conditional_edges(
    "intermediate",
    decide_intermediate_step,
    {
        
        "gen-content":"content",
        "insights":"gen_insights",
        "design":"design",
        "posts":"posts",
        "feedback":"feedback",
        "about-repo":"about_repo",
        "END":END
    }
    
)
graph.add_edge("content","intermediate")
graph.add_edge("gen_insights","intermediate")
graph.add_edge("design","intermediate")
graph.add_edge("posts","intermediate")
graph.add_edge("feedback","intermediate")
graph.add_edge("about_repo","intermediate")
app = graph.compile()
conversational_history = []
if __name__ == "__main__":
    user_inp = input("User: ")
    while user_inp !='exit':
        conversational_history.append(HumanMessage(content=user_inp))
        result = app.invoke({"messages":conversational_history[-3:]})
        conversational_history = result['messages']
        print("AI: ",dict(conversational_history[-1])['content'])
        user_inp = input("User: ")