ar-workspace-safety
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements strict path isolation, limiting all destructive actions (write/delete) to a dedicated workspace directory (
$DATA_DISK/workspace/) while explicitly forbidding access to sensitive system paths such as/etc,/usr,/var, and user SSH configurations. - [SAFE]: It enforces strong credential security by prohibiting the reading, printing, or storage of API keys and private keys in files, requiring them to stay in volatile environment variables only.
- [SAFE]: The skill downloads Miniconda from its official, well-known source (
https://repo.anaconda.com/miniconda/) to ensure a verified and isolated Python runtime, avoiding the risks associated with system-level package managers or third-party repositories. - [SAFE]: It mandates human-in-the-loop confirmation for destructive actions, requiring the agent to list specific targets before performing deletions or repository cleanups.
- [SAFE]: The skill explicitly forbids privilege escalation (sudo), modification of shell startup scripts (
.bashrc), and the use of the system Python interpreter, maintaining a high degree of environment stability and security.
Audit Metadata