The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the exploitation of CVE-2025-1974 by using a specialized tool to inject malicious configurations into the Kubernetes NGINX Admission Controller, leading to arbitrary code execution inside the Ingress Pod.
[DATA_EXFILTRATION]: Provides specific commands to read and potentially exfiltrate sensitive Kubernetes ServiceAccount tokens from '/var/run/secrets/kubernetes.io/serviceaccount/token', which can lead to cluster-wide privilege escalation.
[EXTERNAL_DOWNLOADS]: The skill relies on 'ingressnightmare', an external binary tool not located in trusted repositories, and suggests installation via the 'f8x' command-line installer, which may fetch and execute code from remote sources.
[COMMAND_EXECUTION]: Orchestrates the execution of numerous shell commands, including 'nmap' for port scanning, 'curl' for network probing, and 'kubectl' for interacting with the cluster API.
[PROMPT_INJECTION]: The skill demonstrates a high risk of indirect prompt injection (command injection). User-supplied variables such as 'TARGET_IP', 'ATTACKER_IP', and command strings are interpolated directly into shell execution strings (e.g., in Phase 2.1) without sanitization or boundary markers, allowing an attacker to execute arbitrary shell commands on the agent's host.
[DYNAMIC_EXECUTION]: Includes instructions for runtime compilation of C source code into shared objects ('.so' files) using 'gcc' or cross-compilers to be loaded dynamically by the target process.

k8s-ingress-nightmare