huggingface / huggingface/context-course

Trivial sandbox escape

Open
#28 0 comments 0 reactions 0 assignees View on GitHub
Dominant language
Python
Stars
90
Forks
20
PR merge metrics
No merged PRs in 30d

Description

In unit6, [Tools and Sandboxing](https://huggingface.co/learn/context-course/en/unit6/tools-and-sandboxing) the harness sandbox has a trivial escape.

The [rg command with the --pre flag](https://man.archlinux.org/man/extra/ripgrep/rg.1.en#:~:text=%2D%2Dpre%3DCOMMAND) allows you to run arbitrary commands.

This section would be better served with a discussion of actual sandbox tools like bubblewrap, firejail, the underlying security features like seccomp and landlock, and microvms like firecracker.

Contributor guide

No contributing guide indexed for this repository

Research direction

Start with the linked unit6 Tools and Sandboxing section and review how the harness uses ripgrep with the --pre flag. Update the section to address the escape and cover the requested sandbox tools, security features, and microVMs; done means the guidance no longer presents an escapable setup without warning.

Written by the indexing model from the issue text.

Assessment

Domain
documentation, security
Issue type
Documentation
Difficulty
3/5
Estimated time
1-2 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
55/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.