Stop burning credits on the same bad patterns. Revefi just dropped agent-analyzer, a new GitHub plugin for Claude Code that audits your recent sessions to find repetitive work. The tool identifies tasks your AI agent keeps failing at or repeating, then suggests concrete fixes like new skills, scripts, or CLAUDE.md rules to stop paying for the same mistakes over and over.

How the Privacy-First Analysis Works

The architecture is strictly local-first. The plugin reads your last 14 days or 100 latest sessions, compressing them to under 200 MB. Before anything leaves your machine, it runs a redaction pass on your local hardware. It scrubs API keys, JWTs, emails, phone numbers, and even file contents from tool outputs. You get a preview in .revefi/upload-preview/ showing exactly what was removed. Only after you explicitly approve does the data hit optimizer.revefi.com for analysis.

Installation and Configuration Requirements

You need Node.js 22.12 or newer and Claude Code running in a terminal, IDE, or the desktop app. For terminal users, it’s a two-step install: /plugin marketplace add revefi/agent-analyzer followed by /plugin install revefi@revefi. Desktop users navigate to Settings β†’ Plugins to add the marketplace URL. Once installed, run /revefi:analyze in any repo. The first run requires a Google sign-in or emailed code. Note that the plugin adds a single hook that only triggers after Revefi’s own commands, ensuring it never sends data in the background.

Key Takeaways

  • Redaction is aggressive but not perfect: It catches known formats for secrets and PII but may miss names or addresses, so read the preview before uploading.
  • Fixes are actionable: The report doesn't just list errors; it ranks repeated tasks by spend and proposes specific CLAUDE.md rules or hooks to prevent recurrence.
  • Data sovereignty: Reports are private by default, and you can delete everything you've uploaded via a simple command.

The Bottom Line

If you are running Claude Code in production, you are likely leaking money through inefficient agent loops. This tool closes that gap by turning session logs into actionable configuration updates. It’s a necessary utility for anyone serious about agent cost control.