pr-triage

Inspiration

This project was inspired by a real open-source maintainer working on a project with more than 50,000 GitHub stars and hundreds of forks.

As the project became more popular, the maintainer started receiving dozens, and sometimes hundreds, of pull requests every day. The difficult part was not receiving contributions. The difficult part was understanding which contributions were ready, which ones needed more information, and which ones required immediate attention.

Some pull requests did not follow the project’s required template. Some did not include enough context. Some had failing local checks or CI pipelines. Others were connected to issues already labeled as high priority, but that information was buried across GitHub and difficult to review at scale.

The result was a noisy pull request queue where important fixes could easily get buried under less urgent work.

What is PR Triage?

PR Triage is an AI teammate for open-source maintainers.

A maintainer provides a GitHub repository and a time window. The agent collects the recent open pull requests and builds a complete picture of each one.

It checks whether each PR follows the repository’s own template, separates compliant and non-compliant submissions, classifies valid PRs as bug fixes or features, finds linked issues and priority labels, reviews CI and build checks, and produces a clear report.

The goal is simple: help maintainers understand what needs attention first without manually opening and investigating every pull request.

How it works

PR Triage uses the Strands Agents SDK to coordinate the analysis workflow.

The maintainer selects a repository and time window.
The agent retrieves open pull requests and related GitHub context.
It discovers the repository’s pull request template.
It checks each PR for the required information.
It classifies compliant PRs as bugs or features.
It finds linked issue numbers and priority labels such as P0 or P1.
It reviews available CI and build results.
It returns a Markdown, JSON, or CSV triage report.

Pull requests that do not follow the template are not deleted or hidden. They are placed in a separate section with an explanation, so maintainers can decide what to do next.

Why it matters

Maintainers should spend their time making technical decisions, not repeatedly collecting the same context from hundreds of pull requests.

PR Triage turns a large and noisy GitHub queue into a prioritized view of the work. It helps maintainers find urgent bug fixes faster, identify useful features, spot incomplete submissions, and make review decisions using evidence instead of guesswork.

Built with

AWS Strands Agents SDK
GitHub GraphQL API
Gemini models
Local llama.cpp models
Python
Markdown, JSON, and CSV reporting

The project can also reuse a host-provided Strands model when embedded inside systems such as Hermes or OpenClaw.

Future direction

The next step is to make PR Triage run continuously for large open-source projects. It could monitor new pull requests in the background, notify maintainers only when a high-priority issue appears, and learn the review conventions of each repository over time.