[EXTERNAL_DOWNLOADS]: The documentation and skill instructions recommend installing the tool by downloading a shell script from the author's official GitHub repository (raw.githubusercontent.com/github/gh-aw-firewall) and piping it directly to bash. While this pattern is flagged as a potential risk, it is the intended and standard installation method for a trusted utility produced by 'github'.
[COMMAND_EXECUTION]: The skill requires the use of sudo, docker, and iptables to manipulate host network rules and manage security containers. These high-privilege operations are inherent to the tool's primary purpose of enforcing L7 domain whitelisting and network isolation.
[SAFE]: The skill incorporates significant security hardening features. It uses an LD_PRELOAD shared library ('one-shot-token') to protect sensitive credentials by caching them and immediately unsetting them from the process environment to prevent exfiltration. It also employs tmpfs overlays to prevent the agent from reading sensitive configuration files or logs within the sandbox environment.

awf-skill