Agent-Field/pr-af

What it solves

PR-AF is an open-source agentic code reviewer designed for deep architectural audits rather than shallow diff summaries. It aims to reduce false positives and identify complex, systemic vulnerabilities that linear analysis tools often miss, providing a high-fidelity gate for CI/CD pipelines.

How it works

PR-AF uses a dynamic pipeline that morphs its execution graph based on the specific topology of a Pull Request. The process follows several key phases:

1. Dynamic Pipeline Generation: It evaluates the PR through semantic, mechanical, and systemic lenses to spawn specialized, ephemeral reviewer agents.
2. Evidence Grounding: The system verifies findings by pulling exact caller snippets and import context from the repository to ensure claims are grounded in actual code.
3. Compound Vulnerability Synthesis: It clusters related risks across multiple files to identify larger systemic issues.
4. Falsifiability Gates: Before posting a comment, the system attempts to invalidate the finding based on safe or intended behavior and existing mitigations.

Who it’s for

It is intended for engineering teams who require rigorous, high-accuracy code reviews as a final CI/CD gatekeeper, particularly where architectural depth and the minimization of false positives are more important than execution speed.

Highlights

• Benchmark Leader: Ranked #1 open-source reviewer on the Martian Code-Review-Bench for golden recall.
• Model Flexible: Supports various models, from cheaper models for routine PRs to frontier models like GLM-5.2 or Opus-class models for high-stakes reviews.
• Evidence-Based: Uses programmatic AST extraction to ensure findings are backed by code evidence.
• Cost Effective: Approximately 10x cheaper per review than closed-source commercial tools.
• GitHub Integration: Can be triggered via a GitHub Action using a specific label (pr-af).