The "Self-Evolving" Reviewer

Can an AI Agent Fix Itself? My Weekend Experiment with Recursive Code Reviews

Last week, I was deep into integrating ESP32 support with Claude Code. It was manual, it was specific, and it got me thinking: Why build a one-off fix when we can build a systematic, self-evolving agent?

This weekend, I shifted gears to OpenCode-Review. My goal was to move beyond simple scripts and into a generic "Review & Fix" agent that could be plugged into any task—whether it’s a firmware project on an ESP32 or a complex cloud backend.

The "Dogfooding" Phase: Evolution in Action

The most exciting (and frustrating) part of this journey has been "dogfooding." I wanted the agent to improve its own source code—a digital version of human evolution.

• The First Runs: Total failure. The logic looped, the context was lost, and the agent couldn't see its own flaws.
• The Breakthrough: After three intensive iterations of self-correction, the agent finally started "understanding" its own architectural rough edges. It began fixing issues it had created in earlier versions.

The Wall: Innovation vs. Rate Limits

We are at a point where the software is ready to sprint, but the infrastructure is lagging. Today, progress on OpenCode-Review hit the inevitable ceiling: API Rate Limits. When you’re running a multi-agent system that reviews, tests, and fixes code recursively, you burn through tokens at an incredible rate.

Join the Project: A Call for Infrastructure & Collaboration

I am committed to making OpenCode-Review a robust, stack-aware force multiplier for developers, but I need to accelerate the testing phase.

I’m looking for two types of partners:

1. Collaborators: If you’re working with unique stacks (Embedded, Web, or DevOps) and want to help test a systematic AI reviewer, let’s talk.
2. Infrastructure Support: To bypass the "rate-limit wall" and keep the evolution moving, I’m seeking partners who can provide API credits or Pro environment access (Claude-Code, Cursor, etc.).

If you believe in open-source AI tools that actually understand the code they review, I’d love to connect. Check out the repo, star it, or slide into my DMs to help push this forward.

Repo: https://github.com/devidasjadhav/opencode-review

#AI #OpenSource #ClaudeCode #LLM #EmbeddedLinux #AutonomousAgents #SoftwareEngineering #BuildInPublic