What it does

The Codex Memory Curator is a specialized maintenance utility designed to audit and prune AI agent "memories." As agents accumulate long-term context, they often suffer from memory pollution—stale rules, duplicate entries, or sensitive data leakage across different repositories. This skill provides a structured workflow to inventory, scan for risks, and clean up your agent's durable state.

Why use this skill

Standard prompting can't easily fix an agent's underlying memory architecture. This skill acts as a professional librarian and security auditor for your agent. It classifies every memory entry into one of several categories (KEEP, MOVE TO AGENTS.md, DELETE, etc.), ensuring that repo-specific rules stay in the repo and global preferences stay in memory.

Key advantages include:

• Security First: Employs a redacted risk scanner to identify credentials or sensitive data before you see them.
• Safe Operations: Never performs destructive edits without creating a timestamped backup and receiving explicit confirmation.
• Context Optimization: Shrinks context windows by moving misplaced documentation from memory into the appropriate project files or config.

Supported tools

Developed for developers using the Codex ecosystem, it interacts with ~/.codex/memories and config.toml using specialized Node.js scripts for non-destructive inventory and risk assessment.