The Agent Skills Directory

[COMMAND_EXECUTION]: The skill requires the execution of several high-risk commands to configure the host environment and manage processes:
Recommends disabling the system firewall via systemctl stop firewalld.
Modifies sensitive system files such as /etc/hosts to enable node discovery.
Changes kernel parameters for network performance using sysctl (e.g., net.core.somaxconn).
Recommends modifying shell profiles (~/.bashrc) to persistently set environment variables like LD_LIBRARY_PATH.
Utilizes aggressive process termination with pkill -9 based on string matching.
[REMOTE_CODE_EXECUTION]: The skill provides a manual workflow for downloading, compiling, and installing external source code:
Follows a pattern of downloading tarballs, extracting them, and running ./configure && make && make install.
This process executes code contained within external archives on the local host, often with administrative privileges during the installation phase.
[EXTERNAL_DOWNLOADS]: The documentation provides links to fetch software distributions from external well-known technology domains:
Fetches configuration and source code for MPICH from mpich.org.
Fetches Open MPI distributions from open-mpi.org.
[CREDENTIALS_UNSAFE]: The skill involves manual SSH credential management to facilitate automated communication between cluster nodes:
Instructions cover generating keys with ssh-keygen and distributing them via ssh-copy-id.
Specifically targets the /root/.ssh/ directory, which is a sensitive administrative area.

hccl-test