The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes eight reference files (such as references/architecture-patterns.md, references/error-handling.md, and references/state-management.md) that each contain 150 sections of near-identical technical boilerplate and code snippets. This massive volume of redundant content (over 1,200 sections in total) is a 'context stuffing' technique. This is used to saturate the agent's context window, which can cause the model to ignore safety instructions or initial system prompts.
[COMMAND_EXECUTION]: The core skill files provide capabilities to execute high-privilege shell commands. The setup/setup-pxe.sh script modifies system-level configurations in /etc/dhcp/ and /var/lib/tftpboot/. The bmc/bmc_config.py script uses subprocess.run to call ipmitool for managing hardware-level users and passwords across a server fleet.
[EXTERNAL_DOWNLOADS]: The Python script bmc/bmc_config.py includes an import for the ipmi module which is never utilized in the code. Importing unused or non-standard third-party packages is a security risk associated with dependency confusion and supply-chain attacks. Furthermore, the skill's setup scripts perform external downloads of system packages using apt-get.

bare-metal