The Agent Skills Directory

[COMMAND_EXECUTION]: The provided assessment script executes system commands to enumerate container privileges and runtime versions.
Evidence: The _run function in scripts/agent.py utilizes subprocess.run to call curl for Docker socket checks and runc --version for vulnerability detection.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and execute established security introspection and penetration testing tools from GitHub.
Evidence: SKILL.md contains commands to fetch amicontained, deepce, and CDK from their official GitHub release pages.
[DATA_EXFILTRATION]: The skill documents methods to access sensitive host files as a validation step for container escape testing.
Evidence: Steps 2, 3, and 5 in SKILL.md provide commands intended to read /etc/shadow from the host filesystem to verify isolation bypass.
[REMOTE_CODE_EXECUTION]: The documentation includes proof-of-concept steps for exploiting specific container runtime vulnerabilities (CVEs).
Evidence: Detailed exploitation flows for CVE-2024-21626 and the 2025 runC procfs write-redirect family are provided for educational and authorized testing purposes in SKILL.md and references/api-reference.md.
[INDIRECT_PROMPT_INJECTION]: The assessment script parses output from external binaries, which represents a potential ingestion point for untrusted data if the environment is already compromised.
Evidence: In scripts/agent.py, the check_runtime_cves function uses regex to parse the output of the runc command to determine the version.

escaping-containers-to-host