Gage

Gage

Scans your Claude Code session history for hidden bugs, rule violations and problematic agent behavior

🔗 Visit Gage
📁 Editors, IDEs & Dev Tools🗣️ English📅 September 5, 2026

Description

AI coding agents can quietly do things you'd never approve of if you saw them happen in real time — skip a test, silently work around a broken function instead of fixing it, contradict a project rule you set. Gage is built to go back through the transcript afterward and catch exactly that.

Gage analyzes Claude Code session transcripts to surface unfixed bugs, violated project rules and problematic agent behavior, backed by evidence traced to the original session. It runs four specialized scanners — general, code-review, hidden-thinking and session-retention — and lets you resolve flagged issues interactively with a `gage resolve` command. All data stays local in ~/.gage/, queryable through a full SQL interface, and it plugs into Claude Code as a local MCP server. It's written in Rust with scanners in the Rune language, and currently supports Linux and macOS (Windows is planned).

💬 Our review

The short version: for anyone running Claude Code regularly, retroactively auditing sessions for silently skipped work or rule violations is a genuinely useful safety net that few other tools offer.

There isn't a direct competitor doing exactly this — most code-review tools look at diffs, not the agent's session behavior that produced them. The tradeoffs are narrower applicability (only useful if you're actively using Claude Code, and need a Pro subscription given the token cost of running scanners), no Windows support yet, and being open source with the usual early-project caveats. Worth adding to your workflow if you already run Claude Code at any volume and want visibility into what it quietly did or didn't fix; irrelevant if you don't use Claude Code.

📊 Global score

53Average
🌐Availability15/100Faible

1 language · 0 platform

📄Profile90/100Excellent

Profile completeness

🤖 AI-enriched data

💰 Pricing model
🆓 Gratuit (open source)

Gratuit, licence Apache 2.0. Nécessite un abonnement Claude Code Pro pour l'usage en tokens des scanners.

👥 Target audienceDéveloppeurs utilisant Claude Code régulièrement et voulant auditer le comportement de l'agent
🗣️ Languagesen
🌍 Target countriesWorldwide
👍

Pros

Fonctionne entièrement en local, confidentialité totale des données

Détection d'anomalies appuyée sur des preuves traçables jusqu'à la session d'origine

Interface SQL complète pour explorer les données de session

S'intègre à Claude Code comme serveur MCP local

👎

Cons

Limité à Linux/macOS, le support Windows est seulement prévu

Nécessite un abonnement Claude Code Pro (usage de tokens)

Cas d'usage très spécialisé — utile seulement pour les utilisateurs actifs de Claude Code

❓ Frequently asked questions

What is Gage in one sentence?
Is there a free plan?
Does my data leave my machine?
What platforms does it support?
Is it worth the money compared to alternatives?
Which tool should you pick for your case?