I keep bouncing between Claude Code and Codex while working on different repos. The annoying part is not the AI assistant itself. It is remembering which terminal, project, and conversation belongs to which task after a restart or a long break.
A workflow that has helped me:
- Keep one workspace per repo, not one giant terminal pile.
- Name sessions around the task, not the tool:
fix-auth-tests,refactor-importer,release-notes. - Review old AI sessions before starting a similar task. There is often useful context hiding in previous prompts and patches.
- Treat session recovery as part of the dev environment, like restoring editor tabs.
I have been working on an open-source desktop app around this exact workflow: Shelf.
It is a Tauri app for managing Claude Code and Codex workspaces in one window. The main things it does are:
- group AI coding conversations by project
- reopen previous sessions without remembering session IDs
- scan and organize conversations automatically
- keep a file tree and real PTY terminal next to the session list
- restore workspace state after restarting
It is still early, but active. The current release is v0.2.18 and it supports macOS Apple Silicon and Linux.
The part I care about most is reducing the small bits of friction that make AI coding sessions feel disposable. If the context from yesterday is easy to find, it is much easier to continue the work instead of asking the model to rediscover everything.
Top comments (0)