sandbox-escape

SUSPICIOUS: The skill is internally coherent as a sandbox-escape auditing guide and shows no credential theft, exfiltration, or supply-chain abuse. However, it gives an AI agent explicit offensive security techniques and RCE-oriented escape payloads, which makes it a high-risk security/exploit capability even though the documentation itself is otherwise well-scoped and local-only.

This fragment is an explicit, highly actionable sandbox-escape and host-RCE PoC skeleton. It contains multiple escalation paths from an eval-like API into host process access and then into OS command execution (Node child_process.execSync and Python os.system), with additional evidence collection of sensitive host data (process.env). Any package/version exposing an evaluate-like capability to untrusted input in a vm/sandbox context—directly or indirectly—would represent a severe security risk.