Skip to content

AGENTS.md

Latest commit

 

History

History
51 lines (40 loc) · 2.82 KB

AGENTS.md

File metadata and controls

51 lines (40 loc) · 2.82 KB

AI Agent Instructions

You are an expert Python developer working on this repository. Read and strictly adhere to these rules before writing or modifying any code.

Project Overview

This repository shares tools developed by Jigsaw as a proof of concept to help make sense of large-scale online conversations. It demonstrates how Large Language Models (LLMs) like Gemini can be leveraged for such tasks. The code provided here offers a transparent look into Jigsaw's methods for categorization, summarization, and identifying points of agreement and disagreement in free response public opinion research. Our goal in sharing this is to inspire others by providing a potential starting point and useful elements for those tackling similar challenges.

More details can be found in the README.md file, along with instructions for running each step of the Sensemaking pipeline.

Repository Boundaries & Rules

  • DO NOT touch the case_studies/ directory. Do not read from it, use it for context, or modify any files inside it. This is intended only to document past work, it does not need modification and should not be imported into other files.
  • Source Code Only: All active development happens exclusively within the src/ directory.
  • Google Style Guide: All Python code must strictly follow standard Google Python Style conventions. All functions, classes, and modules must include Google-style docstrings.

Architecture & Tech Stack

1. LLM / Gemini Integration

We use a custom wrapper for all Gemini model interactions.

  • NEVER import google-generativeai, vertexai, litellm, or other direct API libraries in the feature code.
  • ALWAYS use the internal wrapper located at src/models/genai_model.py.
  • (Agent Note: If you are asked to implement a new LLM feature, first review src/models/genai_model.py to understand its expected inputs, outputs, and error handling).

2. Data Handling (CSV)

  • We use pandas exclusively for reading, writing, and manipulating CSV data.
  • Do not use Python's built-in csv module unless explicitly requested.

Executable Commands

Use these exact commands when verifying your work. Run them from the project root.

  • Install dependencies: 
    pip install -r requirements.txt
  • Run all tests: 
    pytest
  • Run a specific test file: 
    pytest path/to/test_file.py
  • Run a specific test function: 
    pytest path/to/test_file.py::test_function_name

Agent Workflow

  1. When asked to create a new file, place it in the appropriate subdirectory within src/.
  2. Before presenting code, write unit tests for it using pytest.
  3. If your code requires new dependencies, ask the user for permission before adding them to requirements.txt.