You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Consider this situation. If, like me, you're a "long term" Linux user, you'll probably been here more times than one:
Upated Ubuntu
Driver/kernel other stuff doesn't work
System looks "bricked" (black screen on boot) but there's a good chance it's fix-able
You have a Zoom meeting, then a pickup run, then work
I recognise that AI agents need to be used cautiously when working on live enivonments. I've seen them brick stuff and delete videos (yup!). But I've also seem them very reliably and very efficiently fix complicated config issues in seconds. It's a risk - decide if it's one you want to take.
Implementation
Traditionally, recovery is done from a shell environment, which is kind of barebones.
This approach isn't Claude-specific (it's just what I'm liking and using at the moment). Any AI CLI that can execute bash commands (Codex, Qwen, Gemini) could work too.
The idea, intended for ease of operation when squatting in a server rack and typing on one arm:
Easy shell installation script that the user pulls with curl at an easy to type URL (maybe with a URL shortener). Example: bit.ly/helpclaude
Curl pulls in a bash script that will get the user set up quickly. Limitation: internet/networking needed and host FS may not yet be RW mounted.
The bash script:
Runs the Claude Code installer (via bash also)
Downloads a couple of text files with slash commands, CLAUDE.md
Downloads a quick user manual like README.txt.
Idea is this:
Slash commands quickly profiles the env and takes a detailed snapshot of the system hardware and what looks broken (ie, X is running but these logs look like there's been Y).
CLAUDE.md is "seeded" with some starter knowledge (this is a recovery environment; see the text file for details)
Slash command (or bash script) should also seed Claude with maximum permissiveness as this is a temporary env: full execution on /!
Self Destruction
Claude works with the user to debug and hopefully fix the environment
Finally, user has an option for how it would like Claude to handle persistence. Now that we have Claude and your system works .... should I stick around?
If "Yes" bash script just deletes the elevated permissions so that Claude is avialable but the permission elevation doesn't persist.
If "no" bash script uninstalls Claude then reboots OS.
To view an index of my Claude Code related projects, click here.
About
WIP/Plan/Idea: System recovery/rescue utility that seeds Claude Code installation for AI-assisted repair from TTYS/shell