The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill contains exhaustive documentation and command sequences designed to bypass container isolation and gain host-level root privileges. It details methods for mounting host devices, entering host namespaces using nsenter, and exploiting Linux capabilities such as CAP_SYS_ADMIN and CAP_DAC_READ_SEARCH.
[COMMAND_EXECUTION]: The instructions include numerous high-risk shell commands for system manipulation, including exploiting cgroup v1 release agents to execute arbitrary commands on the host and using gdb or Python for process injection.
[PERSISTENCE]: The DOCKER_ESCAPE_CHAINS.md file provides explicit instructions for maintaining access to the host after a successful breakout. This includes adding an attacker's public key to the host's /root/.ssh/authorized_keys and setting up a recurring reverse shell backdoor in /etc/crontab.
[DATA_EXFILTRATION]: The playbook specifically targets sensitive system files, providing commands to read the host's /etc/shadow file, which contains encrypted password hashes for all system users.
[CREDENTIALS_UNSAFE]: The skill promotes the unauthorized modification of authentication files (authorized_keys) to bypass standard login procedures and facilitate remote access for an attacker.

container-escape-techniques