Skip to content

Unrestricted LLM-controlled bash execution with zero sandboxing #1

@consigcody94

Description

@consigcody94

Found via code audit. basher.py:243-257. LLM output parsed for bash tags and executed directly as bash script. AGENTS.md in cwd enables persistent prompt injection for arbitrary RCE.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions