What problem does it solve?
Claude Code's personal memory lives in a per-project directory that is locked to Claude Code: it is not version-controlled and is invisible to every other AI CLI. When you run Codex, Cursor, or another tool in the same directory, that tool cannot read your user profile, collaboration preferences, or methodology. This Skill migrates the cross-tool-shareable content out of memory into tool-agnostic reference docs, leaving memory as a thin handoff cache.
Core Features & Use Cases
- Scope-based diagnosis: Classifies every memory file by scope (team rules, cross-tool user profile, temporary handoff) so only the right content migrates, with a privacy check that keeps PII and identity maps out of shared docs.
- Two-layer architecture: Inlines the few must-hold facts into the CLAUDE.md body (guaranteed to load in both tools), moves detail into
~/.claude/references/user/ hub-and-spoke files, symlinks ~/.codex/AGENTS.md to CLAUDE.md, and raises Codex's 32 KiB project_doc_max_bytes truncation limit.
- Multi-agent review and empirical verification: Spawns four parallel review subagents (completeness, cross-references, link integrity, duplication), then verifies by actually running
codex exec and grepping the session rollout log for the inlined section.
- Safe memory cleanup: Cleans, thins, or keeps remaining memory files, greps for dangling cross-references before archiving, and soft-deletes to a dated backup outside
memory/.
- Use Case: You have months of accumulated Claude Code memory and just started using Codex in the same projects. Run this Skill to move your profile and preferences into shared docs, wire the symlink, and prove with a real Codex session that the content actually loads.
Quick Start
Migrate my Claude Code memory into tool-agnostic docs so Codex can read my user profile and preferences too.
