The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads an archive from an arbitrary, user-supplied URL using wget and immediately executes a shell script (bash startserver.sh) contained within the untrusted file.
[COMMAND_EXECUTION]: The skill makes extensive use of the shell to perform system configuration, file management, and process execution.
[PRIVILEGE_ESCALATION]: The instructions require the use of sudo to install Java packages via apt and to modify system firewall rules via ufw.
[PERSISTENCE]: The skill establishes persistence by modifying the user's crontab to ensure a generated backup script runs every hour.
[EXTERNAL_DOWNLOADS]: The skill uses wget to download server modpacks from external, non-whitelisted URLs provided by the user.
[DYNAMIC_EXECUTION]: The skill dynamically generates shell scripts (start-minecraft.sh and backup.sh) using cat and grants them execution permissions via chmod.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by ingesting and inspecting external content.
Ingestion points: Files from the downloaded serverpack.zip are inspected using ls and cat in SKILL.md.
Boundary markers: None; there are no instructions to the agent to treat the downloaded file content as untrusted or to ignore instructions embedded within those files.
Capability inventory: The skill has access to shell execution, administrative privileges (sudo), network downloads (wget), and scheduled tasks (crontab).
Sanitization: No validation, sandboxing, or sanitization is performed on the ZIP contents or the scripts before execution.

minecraft-modpack-server