review-agent-memory

Reads what coding agents have learned and suffered in a project — Claude Code auto-memory, CLAUDE.md/AGENTS.md/GEMINI.md and Cursor/Copilot rules, session transcripts, and context stores such as CONTEXT.md, dual-graph, CodeGraph, and graphify — and turns it into a ranked list of software architecture and harness improvements that cut token usage and agent churn. Mines transcripts with bundled scripts for repeated reads, oversized reads, search loops, failing and retried commands, permission denials, subagent fan-out, and hook-injected context, treats every gotcha memory as a gap the code should close, ranks fixes by estimated tokens saved, writes a report, and applies only what the user picks. Works on one repo or a workspace of several. Use when the user says /review-agent-memory, asks why agents burn so many tokens or keep going in circles in a repo, wants to reduce token usage, agent churn, or context bloat, asks what the agent's memory says about the codebase, wants to turn memories or lessons learned into code or tooling fixes, or asks how to make a project cheaper or easier for agents to work in, even if they don't say 'memory'.

Version

1.0.0