The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads a compressed archive from a user-specified URL and executes a shell script contained within it.
Evidence: ATM10_INSTALL_ONLY=true bash startserver.sh in Step 3, which executes a script extracted from the serverpack.zip downloaded in Step 1.
[PRIVILEGE_ESCALATION]: The instructions command the agent to use elevated privileges to perform system-level changes.
Evidence: Use of sudo apt install to install Java and sudo ufw allow to modify network firewall rules.
[PERSISTENCE]: The skill establishes a recurring mechanism to execute code across user sessions.
Evidence: The configuration of an hourly cron job for backups: (crontab -l ... echo "0 * * * * $HOME/minecraft-server/backup.sh ...") | crontab -.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface by ingesting and executing content from untrusted external sources.
Ingestion points: The <URL> provided for the modpack download in SKILL.md.
Boundary markers: Absent. There are no instructions to verify the integrity or content of the downloaded pack before execution.
Capability inventory: The skill utilizes shell script execution (bash), package installation (sudo apt), firewall modification (sudo ufw), and scheduling (crontab).
Sanitization: Absent. The downloaded archive is processed and its scripts are executed without validation.
[EXTERNAL_DOWNLOADS]: The skill fetches content from external network locations.
Evidence: wget -O serverpack.zip "<URL>".

minecraft-modpack-server