Archive-first therapy for sluggish Codex sessions
A backup-first skill that inspects bloated local Codex state and archives old sessions to keep Codex responsive without erasing context.

What it does This Codex skill is a maintenance script dressed up as a conversational agent. It inspects your local Codex state—SQLite bloat, stale worktrees, large logs, and dead config references—and reports what has grown heavy over time. Only when you explicitly ask it to maintain does it back up and archive old sessions, normalize paths, and rotate logs. By default it is strictly read-only.
The interesting bit The clever hinge is the handoff document. Before archiving a bloated chat, the skill insists you write a small continuity note—current goal, files touched, next steps—so you can kill the heavy thread and resume in a fresh, fast Codex session without losing weeks of context. It treats your chat history like a library that needs pruning, not a landfill that needs burning.
Key highlights
- Report-only by default: runs read-only diagnostics unless you explicitly opt into mutations.
- Archives instead of deletes: moves old sessions, worktrees, and logs to archive folders and writes restore artifacts.
- Targets specific pathologies: oversized thread title/preview metadata in
state_5.sqlite, stale worktrees, largelogs_2.sqlite*files, deadconfig.tomlentries, and Windows path mismatches. - Optional metadata repair: with a specific flag, trims bloated SQLite display titles after backup without touching the actual conversation transcript.
- Respects runtime safety: refuses to mutate local state if Codex is currently running.
Caveats
- The README is mostly a long user manual full of suggested copy-paste prompts; the underlying Python script is secondary.
- It is unclear how the skill actually wires into Codex’s skill system beyond asking Codex to install it.
- Backup folders contain private local metadata, so you must audit them before sharing.
Verdict Worth a look if your local Codex has started lagging after weeks of heavy use and you are paranoid about losing context. If you do not use Codex locally or you already prune your own state by hand, you will not need it.
Frequently asked
- What is vibeforge1111/keep-codex-fast?
- A backup-first skill that inspects bloated local Codex state and archives old sessions to keep Codex responsive without erasing context.
- Is keep-codex-fast open source?
- Yes — vibeforge1111/keep-codex-fast is open source, released under the MIT license.
- What language is keep-codex-fast written in?
- vibeforge1111/keep-codex-fast is primarily written in Python.
- How popular is keep-codex-fast?
- vibeforge1111/keep-codex-fast has 1.5k stars on GitHub.
- Where can I find keep-codex-fast?
- vibeforge1111/keep-codex-fast is on GitHub at https://github.com/vibeforge1111/keep-codex-fast.