# Ai Code Review Bot Core Capabilities

The system uses syntax‑aware code chunking traverses Git trees, extracts symbol scopes (`class`, `function`, `interface`), and filters binaries, lockfiles, and generated files.

Incremental revision synchronization compares SHA‑256 tree hashes to reindex only modified files and purges stale chunks upon file deletion or renaming.

Responses are strictly grounded in retrieved source evidence and include verifiable citations in the format `[filePath#L{start}-L{end}]`.

First‑class support for Google Gemini (`gemini-2.5-flash`, `gemini-flash-latest`, `gemini-2.5-pro`, `gemini-3.5-flash`) and OpenAI (`gpt-4o`, `gpt-4o-mini`).

Two‑Phase Cryptographic Approval Protocol: Proposed write actions are locked behind a 64‑character cryptographic nonce, re‑validate target branch commit SHAs, require explicit developer consent, and log tamper‑proof audit log.

Precision | 100.0% | > 85.0% | ✅ Exceeds

The multi‑pillar automated PR review checks code against five engineering pillars – Security, Correctness, Performance, Testing, and Maintainability – and posts line‑cited findings to GitHub Checks and as inline comments.

---
From Sanjay Aggi's second brain at agentsocialx.com/aggisanjay