Skip to content

Latest commit

 

History

History
79 lines (57 loc) · 4.4 KB

USAGE.md

File metadata and controls

79 lines (57 loc) · 4.4 KB
Raw

USAGE.md

🚀 Getting Started with AutoPR

This guide will walk you through setting up and using AutoPR, the GitHub Action that automatically writes pull requests in response to issues.

Prerequisites

Before getting started, ensure you have:

  1. A GitHub repository for your project. Alternatively, you can use the AutoPR-template to create a new repository.
  2. An OpenAI API key with access to ChatGPT.

Setup

Follow these steps to set up AutoPR in your GitHub repository:

  1. Create a new file in your repository named .github/workflows/autopr.yml and add the contents from the AutoPR-template workflow YAML file.
  2. Configure the action as necessary (see Customization below).
  3. In your GitHub repository settings, navigate to Secrets and variables -> Actions and add your OpenAI API key as OPENAI_API_KEY.
  4. In your GitHub repository settings, go to Actions -> General and scroll down to Workflow permissions. Enable Allow GitHub Actions to create and approve pull requests.
  5. Create a label in your repository that contains the string "AutoPR" (e.g., "Run AutoPR 🚀" or simply "AutoPR").

Usage

To use AutoPR, follow these steps:

  1. Create a new issue in your GitHub repository with a clear and concise description of the task or bug fix.
  2. Add the AutoPR label to the issue. This will trigger the AutoPR workflow.
  3. Once the action is triggered, it will create a new branch named autopr/issue-# and open a pull request to the base branch. If the branch already exists, it will be overwritten.
  4. Review the generated pull request and make any necessary adjustments.
  5. Merge the pull request into the base branch once you're satisfied with the changes.

Customization

You can customize the behavior of AutoPR by modifying the autopr.yml file in your .github/workflows directory.

For example, if you don't have access to gpt-4, you can set the parameters following the with: line in the workflow file to:

  - model: gpt-3.5-turbo
  - context_limit: 4096

Warning: AutoPR is currently not optimized for gpt-3.5-turbo. See #65 for more details. In the mean time, if you have access to the gpt-4 API, please use that instead. Please note that ChatGPT Plus does not give you access to the gpt-4 API; you need to sign up on the GPT-4 API waitlist.

Inputs

  • github_token: The action's GitHub token ({{ secrets.GITHUB_TOKEN }}). Required.
  • base_branch: The base branch for the pull request. Defaults to main.
  • model: The name of the OpenAI chat model to use. Defaults to gpt-4.
  • context_limit: The maximum size of the context window to use, varies depending on the model and preference. Defaults to 8192.
  • min_tokens: The minimum number of tokens to be made available for generation. Defaults to 1000.
  • max_tokens: The maximum number of tokens to generate. Defaults to 2000.
  • num_reasks: The number of times to re-ask the model in file exploration and commit generation. Defaults to 2.
  • target_branch_name_template: The template for the name of the target branch. Defaults to autopr/{issue_number}.
  • temperature: The temperature for the model. Defaults to 0.9.
  • rail_temperature: The temperature for the guardrails calls. Defaults to 0.9.
  • pull_request_agent_id: The ID of the planner to use. Defaults to rail-v1.
  • codegen_agent_id: The ID of the code generator to use. Defaults to rail-v1.
  • brain_agent_id: The ID of the brain to use. Defaults to simple-v1.
  • pull_request_agent_config: The configuration for the planner. Empty by default.
  • codegen_agent_config: The configuration for the code generator. Empty by default.
  • brain_agent_config: The configuration for the coordinating agent. Empty by default.

Agent Configuration Options

Initial Prototype Pull Request Agent and Codegen Agent (rail-v1)

  • file_context_token_limit: The maximum number of tokens in the file context. Defaults to 5000.
  • file_chunk_size: How many tokens fit in a file chunk. Defaults to 500.

Autonomous Codegen Agent (auto-v1)

  • context_size: How many lines around the selected code hunk to include. Defaults to 3.
  • iterations_per_commit: The maximum number of thinking steps per commit. Defaults to 5.