Skip to main content

OpenHands Issue Resolver

Project description

OpenHands Github Backlog Resolver 🙌

Do you have a bunch of open github issues that need to be resolved but no time to do it? What about asking an AI agent to do it for you instead?

This tool allows you to do just that, point towards a github repository, and you can use open-source AI agents based on OpenHands to attempt to resolve issues for you.

It's quite simple to get setup, just follow the instructions below.

Installation

You can install the OpenHands Github Resolver package directly from PyPI using pip:

pip install openhands-resolver

Alternatively, if you want to install from source or contribute to the project, you can clone the repository and use poetry:

git clone github.com/All-Hands-AI/openhands-resolver
cd openhands-resolver
poetry install

If you don't have one already, create a GitHub access token. You can use this link to quickly generate a classic access token. Or, for additional security, you can create a fine-grained token that has "Content" and "Pull requests" scopes for the repository you want to resolve issues in. If you don't have push access to that repo, you can create a fork of the repo and use the fork.

Once you have your token set the GITHUB_TOKEN environment variable, e.g.

export GITHUB_TOKEN="your-secret-token"

You'll also need to have choose an LLM_MODEL and prepare an LLM_API_KEY, for which you can follow the OpenHands setup instructions. OpenHands works best with large, popular models like OpenAI's gpt-4o and Anthropic's Claude.

export LLM_MODEL="anthropic/claude-3-5-sonnet-20240620"
export LLM_API_KEY="sk_test_12345"

Running the agent to resolve issues

After installing the package, you can run the agent to resolve issues using the following command:

python -m openhands_resolver.resolve_issues --repo [OWNER]/[REPO]

For instance, if you want to resolve issues in this repo, you would run:

python -m openhands_resolver.resolve_issues --repo all-hands-ai/openhands-resolver

The output will be written to the output/ directory.

Alternatively, if you only want to resolve a subset of the issues, you can specify a list of issues to resolve. For instance, if you want to resolve issues 100 and 101, you can run:

python -m openhands_resolver.resolve_issues --repo all-hands-ai/openhands-resolver --issue-numbers 100,101

If you've installed the package from source using poetry, you can still use the previous method:

poetry run python openhands_resolver/resolve_issues.py --repo all-hands-ai/openhands-resolver

Visualizing successful PRs

To find successful PRs, you can run the following command:

grep '"success":true' output/output.jsonl | sed 's/.*\("number":[0-9]*\).*//g'

Then you can go through and visualize the ones you'd like.

python -m openhands_resolver.visualize_resolver_output --issue-number ISSUE_NUMBER --vis-method json

Uploading PRs

If you find any PRs that were successful, you can upload them. There are three ways you can upload

  1. branch - upload a branch without creating a PR
  2. draft - create a draft PR
  3. ready - create a non-draft PR that's ready for review
python -m openhands_resolver.send_pull_request --issue-number ISSUE_NUMBER --github-username YOUR_GITHUB_USERNAME --pr-type draft

If you want to upload to a fork, you can do so by specifying the fork-owner.

python -m openhands_resolver.send_pull_request --issue-number ISSUE_NUMBER --github-username YOUR_GITHUB_USERNAME --pr-type draft --fork-owner YOUR_GITHUB_USERNAME

Using the GitHub Actions Workflow

This repository includes a GitHub Actions workflow that can automatically attempt to fix issues labeled with 'fix-me'. To use this workflow in your own repository:

  1. Copy the examples/openhands-resolver.yml file to your repository's .github/workflows/ directory.

  2. Set up the following GitHub secrets in your repository:

    • PAT_TOKEN: A Personal Access Token with repo scope (used for creating PRs and branches)
    • PAT_USERNAME: Your GitHub username
    • LLM_MODEL: The LLM model to use (e.g., "anthropic/claude-3-5-sonnet-20240620")
    • LLM_API_KEY: Your API key for the LLM service
    • LLM_BASE_URL: The base URL for the LLM API (if applicable)

    Note: The workflow also uses the default GITHUB_TOKEN secret for some operations, which is automatically provided by GitHub Actions.

  3. To trigger the workflow, add the 'fix-me' label to any issue you want the AI to attempt to resolve.

The workflow will:

  • Attempt to resolve the issue using the OpenHands resolver
  • Create a draft PR if successful, or push a branch if unsuccessful
  • Comment on the issue with the results

Note: This workflow requires the openhands-resolver package, which will be installed automatically during the workflow execution.

Troubleshooting

If you have any issues, please open an issue on this github repo, we're happy to help! Alternatively, you can join the OpenHands Slack workspace and ask there.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

openhands_resolver-0.1.6.tar.gz (185.0 kB view details)

Uploaded Source

Built Distribution

openhands_resolver-0.1.6-py3-none-any.whl (190.8 kB view details)

Uploaded Python 3

File details

Details for the file openhands_resolver-0.1.6.tar.gz.

File metadata

  • Download URL: openhands_resolver-0.1.6.tar.gz
  • Upload date:
  • Size: 185.0 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: poetry/1.8.3 CPython/3.11.9 Darwin/23.1.0

File hashes

Hashes for openhands_resolver-0.1.6.tar.gz
Algorithm Hash digest
SHA256 fec1ebfcdb2c7fdda3b73e534cf4d0b168371929e1c530a140e7381ea1bf97ae
MD5 9c81db5b353f9c7414493bdde1aad6d2
BLAKE2b-256 acfa6fa5f6ad88f8b5a763d242706ac5531592ba517902e32d1a7f33ebbecf4e

See more details on using hashes here.

File details

Details for the file openhands_resolver-0.1.6-py3-none-any.whl.

File metadata

File hashes

Hashes for openhands_resolver-0.1.6-py3-none-any.whl
Algorithm Hash digest
SHA256 a113e24aa8c7c0e12968c0ab071e209e61dd59671bd45c685c04c4630705602e
MD5 208aa1412349ae493fbfcb31c362a0bc
BLAKE2b-256 26e69f2b229f95e862fcc93188f9f7cd880197bd4e251ac4420fa964d6d871c3

See more details on using hashes here.

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page