pr-af  by Agent-Field

Agentic code reviewer for deep CI/CD audits

Created 5 months ago
323 stars

Top 83.9% on SourcePulse

GitHubView on GitHub
Project Summary

Summary

PR-AF is an open-source, agentic code reviewer designed for deep architectural audits within CI/CD pipelines. It addresses the limitations of shallow diff summaries by employing a dynamic, multi-agent approach to identify complex issues with high accuracy and cost-effectiveness. Targeted at developers and teams requiring robust code quality gates, it leverages advanced LLMs to achieve superior performance on code review benchmarks.

How It Works

PR-AF utilizes a dynamic, morphing execution graph based on the incoming Pull Request's topology. It compiles "review dimensions" (semantic, mechanical, systemic) to spawn specialized, ephemeral reviewer agents. Key innovations include "Evidence Grounding," which verifies findings against exact code snippets and context, and "Compound Vulnerability Synthesis," clustering related risks across files to detect systemic issues. This cognitive pipeline aims for high-fidelity reviews with extremely low false positives, surpassing traditional linear analysis.

Quick Start & Requirements

  • Installation: Install via AgentField CLI (af install https://github.com/Agent-Field/pr-af) or deploy using Docker Compose locally or via Railway.
  • Prerequisites: Requires OPENROUTER_API_KEY for LLM access and a GH_TOKEN (GitHub personal access token with repo scope) for interacting with GitHub.
  • Deployment: Options include one-click Railway deployment, local Docker Compose setup, or integration into GitHub Actions workflows.
  • Documentation: Core concepts detailed in docs/ARCHITECTURE.md.

Highlighted Details

  • Ranked #1 open-source code reviewer on Martian Code-Review-Bench with 0.706 golden recall.
  • Identifies approximately 3x more independently valid findings than leading commercial tools.
  • Offers ~10x cost savings per review compared to closed-source alternatives.
  • Supports flexible model selection, from DeepSeek-class for routine checks to Opus-class for critical reviews.
  • Posts evidence-grounded inline GitHub review comments.

Maintenance & Community

The project is part of the broader AgentField ecosystem. Specific community channels (Discord/Slack) or detailed maintenance roadmaps are not explicitly detailed in the provided README.

Licensing & Compatibility

Licensed under the permissive Apache 2.0 license, allowing for commercial use and integration into closed-source projects.

Limitations & Caveats

Review execution times range from 35-50 minutes, making PR-AF best suited for CI/CD gating rather than fast, interactive development loops. It relies on external API keys for LLM access and GitHub integration.

Health Check
Last Commit

6 days ago

Responsiveness

Inactive

Pull Requests (30d)
1
Issues (30d)
0
Star History
13 stars in the last 30 days

Explore Similar Projects

Starred by Shizhe Diao Shizhe Diao(Author of LMFlow; Research Scientist at NVIDIA), Michael Chiang Michael Chiang(Cofounder of Ollama), and
7 more.

openbench by groq

0%
789
Provider-agnostic LLM evaluation infrastructure
Created 1 year ago
Updated 1 week ago
Feedback? Help us improve.