cirolini/genai-code-review

GitHub Action that reviews pull request diffs with an LLM and posts the comments on the PR

What it solves

This project provides a GitHub Action that uses Large Language Models (LLMs) to automatically review pull request diffs and post inline comments. It specifically addresses the problem of "review fatigue" by implementing a comment budget, ensuring that only the most critical findings are highlighted to the reviewers, rather than overwhelming them with a high volume of low-value comments.

How it works

  1. Filtering: It first removes files matching ignore_paths (such as lockfiles or generated code) before sending data to the LLM.
  2. Incremental Review: On subsequent pushes, it only analyzes the new commits since the last review.
  3. Chunking: The diff is split into chunks to fit the model's context window; any content that cannot fit is reported in the summary.
  4. Schema-based Analysis: The model generates findings based on a JSON schema (including severity, confidence, and rationale). If the output doesn't validate, the tool performs one repair attempt.
  5. Validation: Findings are checked against the diff to ensure they can be placed correctly on GitHub. Invalid placements are moved to the summary.
  6. Posting: The top max_comments (ranked by severity and confidence) are posted as a grouped review, and a sticky summary comment is updated.

Who it’s for

Developers and teams using GitHub who want AI-assisted code reviews that are focused and high-signal, and who use providers like OpenAI, Anthropic, Google Gemini, or self-hosted OpenAI-compatible servers (e.g., Ollama, vLLM).

Highlights

  • Comment Budget: Prevents reviewer fatigue by limiting the number of inline comments and explicitly listing suppressed findings in the summary.
  • Broad Provider Support: Works with Gemini, OpenAI, Anthropic, and any OpenAI-compatible API.
  • Security-First Design: Treats diffs as untrusted data using sentinels to prevent prompt injection and avoids using pull_request_target for fork PRs.
  • Zero-Disk Footprint: Reads pull requests via API without needing actions/checkout.
  • Transparent Reporting: The summary comment explicitly states what was not reviewed (e.g., truncated diffs or ignored files).

Related

  • Project
    DocAILab/XRAG

    XRAG is an open‑source Python framework that benchmarks the individual components of Retrieval‑Augmented Generation (RAG) pipelines. It offers modular plug‑ins for retrievers, embeddings, LLMs, and orchestrators, supports traditional IR metrics plus LLM‑based quality scores, provides a CLI (`xrag-cli`) and a Web UI, and can run as a REST service. Install via PyPI, configure with `config.toml`, and use commands like `xrag-cli run`, `xrag-cli webui`, or `xrag-cli api` to evaluate or serve RAG systems.

  • Project
    microsoft/genaiops-promptflow-template

    An LLMOps template for Prompt flow that streamlines the end-to-end lifecycle of LLM applications, from local experimentation and evaluation to production deployment.

  • Project
    gianlucasb/hallucinator

    A tool to detect fake citations in academic papers by extracting references from PDFs and validating them against multiple academic databases.

  • Project
    intel/opencl-intercept-layer

    A tool for intercepting and modifying OpenCL calls to enable debugging and performance analysis without requiring application or driver modifications.