Session Atlas

Codex and Claude Code both store their conversation transcripts globally — keyed by the working directory a session ran in, not by the repo. That makes it hard to answer simple questions like "which of these worktrees still has a live session" or "where's that conversation where I refactored the auth flow". Session Atlas scans both stores and reassembles that picture for one project at a time: every session, the worktree it ran in, whether it's active or archived, and a summary of what each one actually did.

The work is deterministic — a local server reads the stores and extracts everything live, so there's nothing for you to assemble by hand. Your job is just to launch it.

Launching the UI

The server lives in this skill's own app/ directory. Install dependencies once, then start it — point npm at that directory wherever this skill is installed: