The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses 'Autopilot' instructions, directing the agent to run shell commands and write files without requesting permission at each step. This specifically attempts to bypass standard agent confirmation safety protocols.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-supplied data (domain names, email addresses, and working directories) and interpolates them directly into shell commands and system files without instructions for sanitization or validation.
Ingestion points: User input collected during Phase 2 and 3 (domain, email, and working directory).
Boundary markers: Absent; the skill proceeds to execute commands immediately after receiving input.
Capability inventory: Execution of docker compose, openssl, dig, curl, and crontab across multiple workflow phases.
Sanitization: Absent; inputs are used directly in command strings and configuration file templates.
[PERSISTENCE]: The skill automatically installs a cron job in the user's crontab to handle certificate renewals twice daily. While required for the skill's functionality, this modification to the system's scheduled tasks establishes persistence.
[COMMAND_EXECUTION]: The skill performs extensive shell operations including environment script generation, file permission modifications (chmod +x), and system-level networking checks (dig).
[EXTERNAL_DOWNLOADS]: The skill fetches Docker and Nginx configuration files from the official Phase Console GitHub repository to set up the local environment.

docker-compose