tools.py

import sys
import io
import requests
from bs4 import BeautifulSoup
import pandas as pd
from pypdf import PdfReader
from langchain_core.tools import tool
from langchain_community.tools import DuckDuckGoSearchRun
import pptx
import docx  # If missing, run: pip install python-docx

# 1. Native LangChain Search Tool
web_search = DuckDuckGoSearchRun(
    description=(
        "A wrapper around DuckDuckGo Search. Useful for searching the web for current facts, "
        "academic papers, and conference proceedings. When looking for specific journal articles "
        "like Nature Scientific Reports 2012, do not search full complex sentences. Instead, "
        "pass clean, broad keyword queries (e.g., 'Nature Scientific Reports 2012 conference proceedings nano compound') "
        "to get the best results. If this tool returns no relevant text, immediately fall back to "
        "using `execute_python_code` to fetch resources programmatically."
    )
)

# 2. Resilient Web Scraping Tool
@tool
def fetch_webpage_content(url: str) -> str:
    """Useful when you need to read the full text content of a specific URL webpage. 
    It strips away HTML tags and returns clean text snippets."""
    try:
        headers = {"User-Agent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64)"}
        # Enforce strict 10s connect and read timeouts
        response = requests.get(url, headers=headers, timeout=(5, 10), stream=False)
        response.raise_for_status()
        
        soup = BeautifulSoup(response.text, "html.parser")
        
        # Remove script and style elements
        for script in soup(["script", "style"]):
            script.decompose()
            
        text = soup.get_text(separator=" ")
        lines = (line.strip() for line in text.splitlines())
        chunks = (phrase.strip() for line in lines for phrase in line.split("  "))
        clean_text = "\n".join(chunk for chunk in chunks if chunk)
        
        return clean_text[:6000]
    except Exception as e:
        return f"Error reading page content: {e}"

# 3. Robust Code Sandbox (Standard Output Redirected)
@tool
def execute_python_code(code: str) -> str:
    """Useful to run complex calculations, logical processing, or data manipulation.
    Write full code blocks using print() statements to view target evaluation values."""
    old_stdout = sys.stdout
    redirected_output = sys.stdout = io.StringIO()
    local_vars = {}
    try:
        # Execute code in a clean workspace context
        exec(code, {"pd": pd, "openpyxl": openpyxl if 'openpyxl' in sys.modules else None}, local_vars)
        sys.stdout = old_stdout
        captured_out = redirected_output.getvalue()
        
        if not captured_out and local_vars:
            return f"Execution succeeded. Captured workspace variables: {str(local_vars)}"
        return captured_out if captured_out else "Execution complete with no output print statements."
    except Exception as e:
        sys.stdout = old_stdout
        return f"Execution Error: {e}"

# 4. Specialized Document Analysis Tools
@tool
def read_local_pdf(file_path: str) -> str:
    """Extracts raw string text layout from a target local PDF file document."""
    try:
        reader = PdfReader(file_path)
        extracted_text = ""
        for page in reader.pages[:10]: # Safe chunk limits
            extracted_text += page.extract_text() + "\n"
        return extracted_text[:6000]
    except Exception as e:
        return f"Failed to parse PDF document: {e}"

@tool
def inspect_excel_sheets(file_path: str) -> str:
    """Reads names of all tabs and displays preview slices of spreadsheets for data inspection."""
    try:
        # Read file bytes into memory first to avoid thread deadlocks on Windows
        with open(file_path, "rb") as f:
            file_bytes = io.BytesIO(f.read())
            
        xl = pd.ExcelFile(file_bytes)
        summary = f"Available sheets/tabs: {xl.sheet_names}\n\n"
        for sheet in xl.sheet_names[:3]:
            df = pd.read_excel(file_bytes, sheet_name=sheet)
            summary += f"--- Sheet: {sheet} (Shape: {df.shape}) ---\n"
            summary += df.head(5).to_string() + "\n\n"
        return summary[:6000]
    except Exception as e:
        return f"Failed to analyze spreadsheet structure: {e}"

@tool
def read_local_docx(file_path: str) -> str:
    """Useful to extract raw text paragraph contents layout from a target local Word document (.docx)."""
    try:
        # Read file bytes completely first to prevent thread deadlock issues on Windows systems
        with open(file_path, "rb") as f:
            file_bytes = io.BytesIO(f.read())
            
        doc = docx.Document(file_bytes)
        full_text = [para.text for para in doc.paragraphs if para.text]
        return "\n".join(full_text)[:6000]
    except Exception as e:
        return f"Failed to parse Word document: {e}"
    
@tool
def read_local_pptx(file_path: str) -> str:
    """Useful to extract raw text slide-by-slide from a target PowerPoint presentation (.pptx)."""
    try:
        with open(file_path, "rb") as f:
            file_bytes = io.BytesIO(f.read())
        prs = pptx.Presentation(file_bytes)
        summary = ""
        for i, slide in enumerate(prs.slides, start=1):
            slide_text = []
            for shape in slide.shapes:
                if shape.has_text_frame:
                    slide_text.append(shape.text_frame.text.strip())
            text_content = " | ".join(t for t in slide_text if t)
            summary += f"Slide {i}: {text_content}\n"
        return summary[:6000]
    except Exception as e:
        return f"Failed to parse PowerPoint file: {e}"
    

# Expose complete toolkit array to LangGraph agent executor
all_tools = [
    web_search, 
    fetch_webpage_content, 
    execute_python_code, 
    read_local_pdf, 
    inspect_excel_sheets,
    read_local_docx,
    read_local_pptx
]