The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill pipes raw output from remote target endpoints (Kubernetes pods list and etcd keys) directly into the Python interpreter using python3 -m json.tool and custom Python scriptlets (python3 -c '...'). This pattern, while used here for formatting and parsing, allows data from an untrusted remote target to be processed by a local interpreter, which is a potential injection surface.
[COMMAND_EXECUTION]: The skill relies heavily on the execution of shell commands, including nmap, curl, kubectl, and etcdctl, to perform its intended security auditing functions.
[EXTERNAL_DOWNLOADS]: The skill initiates network connections to external IP addresses and domains defined by the user as targets, as well as cloud-specific metadata services (IMDS) for AWS, Azure, and Google Cloud.
[DATA_EXFILTRATION]: The skill explicitly instructs the user/agent to use Out-Of-Band (OOB) techniques for vulnerability confirmation, such as sending captured tokens or system information to external listener services like Burp Collaborator or interactsh.
[PROMPT_INJECTION]: The skill processes untrusted data (secrets, pod logs, and API responses) retrieved from external targets. It lacks explicit boundary markers or sanitization logic, creating a surface for indirect prompt injection where a malicious target could supply data that the agent might misinterpret as instructions.

hunt-k8s