Skip to main content

codex-settings-sync

Use when the user wants to migrate Codex state in ~/.codex into Agent Filesystem and mount the same shared Codex memory/settings across multiple computers. Recommends a .afsignore before migration and defaults to excluding worktrees, caches, logs, and temporary files.

Datos de origen

Repositorio
redis/agent-filesystem
Última actividad en el origen
1 de mayo de 2026 a las 14:40
Idioma detectado de SKILL.md
inglés
Estrellas
55
Forks
10

Opciones de instalación

De forma predeterminada está seleccionado el prompt que primero revisa el origen. Puedes cambiar a un comando directo o descargar una copia local.

Revisa los archivos de origen

Lee SKILL.md y los archivos complementarios que muestra SkillsMP antes de decidir si quieres instalarlo.

Explorador de archivos
2 archivos

Mostrando SKILL.md

SKILL.md
Instrucciones de origen · Vista previa de solo lectura
name
codex-settings-sync
description
Use when the user wants to migrate Codex state in ~/.codex into Agent Filesystem and mount the same shared Codex memory/settings across multiple computers. Recommends a .afsignore before migration and defaults to excluding worktrees, caches, logs, and temporary files.
# Codex Settings Sync Use this skill when the goal is to share Codex state across machines by moving `~/.codex` into Agent Filesystem, then mounting that same Redis-backed volume on other computers. ## Default stance - Recommend a root `~/.codex/.afsignore` before migration. - Default to excluding `worktrees/`. - Treat `cache/`, `tmp/`, `logs/`, `*.log`, `*.tmp`, `*.pid`, and `*.sock` as good default exclusions. - If the user appears to care about restoring local checkout state inside Codex, call out `worktrees/` as a choice point before migrating. Open the bundled starter ignore file at [assets/.afsignore](assets/.afsignore) and adapt it to the user's needs. ## Migration workflow 1. Ask the user to stop Codex on the machines involved, or verify that it is already closed. 2. Ensure `agent-filesystem` is built with `make`. 3. Configure `afs` to point at the shared Redis instance. 4. Create or update `~/.codex/.afsignore` before migration. 5. On the source machine, run `./afs ws import --mount-at-source .codex ~/.codex`. 6. Explain that the imported workspace is mounted at `~/.codex`, and the workspace name is `.codex`. 7. On each additional machine, move aside any existing `~/.codex`, choose mount mode with `./afs config set --mode mount` if you want a live mount there, then run `./afs ws mount .codex ~/.codex`. 8. Verify with `./afs status` and `ls -la ~/.codex`. ## Secondary machine config Point the CLI at the same control plane or Redis database. Then run `./afs config set --mode mount` for live mount mode, or `./afs config set --mode sync` for sync mode, before mounting `.codex` at that machine's `~/.codex`. ## Notes to surface - `afs ws import` honors `~/.codex/.afsignore` if present. - `.afsignore` uses `.gitignore`-style pattern syntax. - Excluding a directory like `worktrees/` is usually safer than syncing it. - Avoid using the same shared `~/.codex` from multiple active computers at the same time. - Keep any `.local-backup` directories until the setup is stable. ## Rollback Source machine rollback: ```bash ./afs ws unmount ~/.codex ``` Secondary machine rollback: ```bash ./afs ws unmount ~/.codex rm -rf ~/.codex mv ~/.codex.local-backup ~/.codex ```
Ver en GitHub