Every engineering team has felt this pain: a code review bot gets installed with excitement, then quietly disabled within a month because it drowns pull requests in false positives and vague, unverifiable warnings. We wanted to fix the actual root cause not "make a smarter bot," but make one that never speaks unless it can prove what it's saying. The idea crystallized around a simple rule: if the agent can't show a taint trace, a reproduction input, and a concrete fix, it stays silent. That constraint became the entire product.
What it does
ProofReview is an autonomous code review agent that analyzes pull requests the way a senior engineer would not by pattern-matching a diff, but by tracing how data actually moves through the changed code. For every pull request, it:
Reads the diff and maps the surrounding code context Investigates candidate issues across security, performance, and stability categories Verifies each candidate against real evidence before reporting it suppressing anything it can't back up Publishes inline findings with exact file/line references, an evidence trail (taint trace + reproduction input), and a suggested fix Generates an overall PR Risk Score so teams can triage at a glance Supports multiple languages Python, JavaScript, and Ruby through a flexible file-upload and GitHub PR intake flow
Built With
- codex
- environment-configuration
- fastapi
- gpt-5.6
- html/css
- javascript
- openai
- openai's-coding-agent
- orchestrated-through-a-context-gathering-?-investigation-?-verification-pipeline-so-that-every-published-finding-survives-a-self-check-before-it-reaches-the-user.-codex-was-used-throughout-for-building-out-github-pr-integration-(via-pygithub)
- pygithub
- python
- render
- ruby
- the-multi-language-file-upload-flow
- to-scaffold-and-iterate-on-the-entire-fastapi-backend-?-from-the-initial-project-structure-to-the-multi-agent-analysis-pipeline.-the-core-reasoning-engine-runs-on-gpt-5.6
- uvicorn
Log in or sign up for Devpost to join the conversation.