The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs diagnostic commands such as process checks and network probes inside the deployed containers using the run_once action. These executions are scoped strictly to the managed container environments for health verification purposes.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external data from hosting requests and project documentation. While this creates a potential surface for indirect injection, the risk is addressed through the use of a rigid research subagent protocol and a strictly defined schema for processing deployment requirements, rather than direct instruction execution.
[DATA_EXPOSURE]: Management of sensitive configuration data follows a secure pattern. The skill explicitly forbids the agent from handling actual secret values, instead requiring the use of REPLACE_ME placeholders and directing the human user to perform final secret entry via the official platform UI.
[DYNAMIC_EXECUTION]: The skill dynamically assembles Docker Compose configurations and performs base64 encoding on application labels to meet Coolify API requirements. These operations are essential to its function as a deployment orchestrator and do not involve executing untrusted code on the host system.

host