Gage Scans Claude Sessions for Bugs Developers Miss

A new open-source tool called Gage is addressing a critical gap in AI-assisted development workflows: bugs that developers approve without reading Claude's warnings. Built in Rust and released under Apache 2 license, Gage scans Claude Code session transcripts to surface issues that traditional code review misses. The creator reports that the tool caught a real bug within 24 hours of use, a breaking change that Claude had explicitly flagged but the developer approved anyway.

Why Session Transcripts Matter More Than Diffs

Traditional code review focuses on the final diff, showing what changed but not why. Claude Code sessions capture both the code modifications and the reasoning behind each decision, including warnings, alternative approaches, and risk assessments. When a developer says "go ahead" without reading carefully, that context disappears from the commit history. The only record of Claude's warning exists in the session transcript.

Gage treats these transcripts as a primary source for post-session audit. Every issue the tool raises must cite specific lines in the session where Claude flagged a concern or made a questionable decision. This evidence-based approach helps developers understand not just that a bug exists, but where in the conversation they missed the signal.

How Gage Works

The tool operates as both a command-line interface and a terminal user interface. Developers point Gage at their Claude Code session files, and it analyzes the full conversation history. The scanner looks for patterns where:

• Claude explicitly warns about breaking changes or edge cases
• The developer approves without asking clarifying questions
• Code decisions contradict earlier statements in the same session
• Important context from earlier messages gets ignored in later changes

Because Gage is written in Rust, it processes large session files quickly. The TUI mode provides an interactive view of flagged issues, allowing developers to jump between the problematic code and the relevant session context.

Release Date and Availability

Gage was officially launched on January 2025 via a Show HN post on Hacker News. The source code is available now on GitHub under the Apache 2 license, making it free for commercial and personal use. The project is under active development, with the creator actively soliciting feedback from early users.

Real-World Impact

The creator's reported bug catch demonstrates the tool's practical value. In that case, Claude flagged a risk during the session, the developer approved the change without careful review, and the feature broke in production. The commit diff showed nothing suspicious, just clean code. Only the session transcript revealed that Claude had warned about the exact failure mode that occurred.

This scenario highlights a broader challenge in AI-assisted development: the speed of code generation can outpace human review capacity. Developers working with Claude Code or similar tools may approve dozens of changes in a single session. Gage provides a safety net by automatically reviewing what the developer might have skimmed.

What This Means

Gage represents a new category of developer tooling: post-session audit systems for AI coding assistants. As more teams adopt tools like Claude Sonnet for production work, the gap between AI-generated code speed and human review quality becomes a reliability risk. Gage's evidence-based approach, requiring citations from the actual session where issues were flagged, offers a practical solution. The tool's open-source status and Rust performance characteristics position it well for integration into CI/CD pipelines, potentially becoming a standard quality gate for AI-assisted development workflows.