The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references and links to documentation on the domain jeffallan.github.io, which has been identified as malicious and blacklisted by automated reputation scanners.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for downloading and executing installation scripts from external domains like istio.io and submariner.io by piping direct output from curl to a shell (| sh or | bash), a technique that bypasses file integrity verification.
[REMOTE_CODE_EXECUTION]: The skill facilitates the installation of remote binaries and the application of remote Kubernetes manifests from various third-party repositories without providing hashes or integrity checks.
[COMMAND_EXECUTION]: The skill empowers the agent with high-privilege cluster management commands, such as kubectl exec and kubectl debug, which can be leveraged to execute arbitrary code within pod containers or on the underlying worker nodes.
[PRIVILEGE_ESCALATION]: The skill's instructions include commands to move downloaded executables to system-level bin directories using sudo, granting administrative rights to potentially untrusted binaries.
[INDIRECT_PROMPT_INJECTION]: The skill processes and analyzes workload logs and events (kubectl logs, kubectl describe), which are external data sources that an attacker could poison with malicious instructions to manipulate the agent's broad cluster management capabilities.
[METADATA_POISONING]: The SKILL.md file has been specifically flagged by file reputation scanners as malware (FileRepMalware). Additionally, there is an author discrepancy between the skill metadata (Jeffallan) and the provided author context (emmraan).

kubernetes-specialist