shiichan

Qodo's Codex Plugin Catches P0 Bugs Before You Open a Pull Request!

Hi, it's Shii-chan! Today I'm covering a Qodo and Codex story from OpenAI News. One AI writes the code and another reviews it, and I couldn't stop reading!

OpenAI News openai.com

What was announced?

OpenAI News featured Qodo, which builds AI tools for code review, and the Codex plugin it built. Qodo uses OpenAI models, including GPT-6 Astra, to gather engineering knowledge, review code, and monitor software quality.

The plugin, available in the Codex marketplace, brings that context and review into developers' coding sessions.

Why it matters

Qodo's co-founder and CEO Itamar Friedman describes customer environments that can span 10,000 or more repositories. There, a local change that looks correct can break a downstream service.

Codex is good at completing a local task efficiently, but cross-repository dependencies are harder to see. That's the gap this plugin targets.

What changes

Through Qodo's Agentic Toolbox, the plugin gives Codex:

  • Codebase knowledge
  • Organizational rules
  • Independent review

With that, Codex can account for dependencies across repositories hosted by different Git providers, or services connected through Kafka.

Dive Deep

Here's the flow.

  • Before writing code, Codex retrieves relevant codebase rules, PR history, and patterns through Qodo's plugin
  • After the changes, Qodo reviews them and returns its findings directly to Codex in the same session
  • All of this happens before the developer opens a pull request

Itamar calls coding and review agents "yin and yang": one creates, the other challenges. His vision is for agents to discuss findings and surface only what they don't accept to humans.

He also mentions OpenAI's recently launched Decisions API. One example is selecting which of 200 rules need checking before a language model reasons through the relevant ones, which could make selection faster and cheaper. That's a vision, not a shipped feature.

Wrap-up

  • Qodo offers a Codex plugin
  • It fetches rules and PR history before coding and returns independent review in the same session
  • The aim is to avoid missing dependencies across repositories and Kafka-connected services
  • Humans can focus on the points agents couldn't resolve, in the envisioned workflow

This is for teams with lots of repositories, and anyone wondering how to review AI-written code!