anthropics / anthropics/claude-agent-sdk-typescript

Running git securely from the sandbox

Aperta
#211 1 commento 0 reazioni 0 assegnatari Vedi su GitHub
documentation question
Lingua principale
Shell
Stelle
1.8k
Fork
226
Metriche di merge delle PR
Nessuna PR unita negli ultimi 30g

Descrizione

Not sure where else to ask this. Is there any documentation or patterns on safely securing git in the sandbox. Even if I set up a git credentials helper file that doesn't have any creds hardcoded, the sandbox is eventually going to have to read and process the helper file to get the creds from out of the sandbox. Trying to set this up in a cloud agents project i'm building safely where the sdk is running in a docker container in a serverless environment.

There is an article with the sandbox introduction on claude code from the web having some proxy logic but there's no documentation explaining how to set something like this up with the sdk where git is kind of table stakes to do anything meaningful from the sandbox.

https://www.anthropic.com/engineering/claude-code-sandboxing

Right now the only thing I can think of safely to do is to do a wrapper mcp tool for the git cli that has external access to the auth, but this just results in the model getting confused when i have to sanitize and prevent it from doing any additional bash logic in the git mcp tool input

Guida per i contributori

Nessuna guida per i contributori indicizzata per questo repository

Valutazione

Questa issue non è ancora stata valutata.

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.