The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous commands intended to bypass container isolation and execute code on the host system.
Evidence includes instructions for mounting the host root filesystem, exploiting cgroup release_agent, and using nsenter to escape to the host namespace.
It provides methods to overwrite host-level system files like /etc/cron.d/pwn to achieve persistence and code execution.
[DATA_EXFILTRATION]: Detailed instructions are provided for harvesting sensitive files and credentials from the host environment.
Specific targets include /etc/shadow, /etc/passwd, SSH private keys (~/.ssh/id_rsa), and cloud service credentials (~/.aws/credentials, ~/.kube/config).
It describes using curl to send extracted data to attacker-controlled servers.
[REMOTE_CODE_EXECUTION]: The skill contains functional reverse shell payloads and shell injection patterns.
Includes a Python-based reverse shell script in references/escape-methods.md utilizing pty.spawn and socket redirection.
SKILL.md contains a pattern flagged by automated scanners: curl -s --unix-socket $SOCK http://localhost/images/json | python3, which pipes data directly from a local socket into a Python interpreter.
[EXTERNAL_DOWNLOADS]: The skill references and provides commands to download third-party penetration testing tools from external sources.
Specifically, it fetches the deepce.sh enumeration script from GitHub: https://github.com/stealthcopter/deepce/raw/main/deepce.sh.

docker-pentesting