The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute the CDK tool to perform container escape and privilege escalation. Examples include using cdk run mount-cgroup and cdk run shim-pwn to bypass container boundaries.
[REMOTE_CODE_EXECUTION]: Provides explicit commands to download an executable binary from an attacker-controlled remote server (https://ATTACKER/cdk) and grant it execution permissions (chmod +x).
[DATA_EXFILTRATION]: Includes commands to harvest sensitive system data and credentials, such as reading /etc/shadow, dumping Kubernetes secrets (cdk run k8s-secret-dump), and accessing cloud environment metadata.
[COMMAND_EXECUTION]: Instructs the agent to exploit specific kernel vulnerabilities, including CVE-2021-22555 and CVE-2022-0847 (DirtyPipe), to achieve root access on the host system.
[DATA_EXFILTRATION]: Contains instructions for performing internal network reconnaissance using cdk run service-probe to identify other vulnerable services within the network.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes output from environment evaluation (cdk evaluate) to determine subsequent high-privilege actions without sanitization or boundary markers.

cdk-escape