The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill documentation provides examples for reading highly sensitive files from the local filesystem, such as ~/.ssh/id_rsa and github-app-key.pem, to be passed as arguments to the deployment platform.
[COMMAND_EXECUTION]: The skill extensively uses a local bash script located at {baseDir}/scripts/coolify. These scripts are invoked with complex arguments (e.g., --json, --value), which may lead to shell command injection if the agent interpolates untrusted user input without strict validation.
[DATA_EXFILTRATION]: The skill is designed to transmit sensitive data, including server private keys and application environment variables, to remote Coolify API endpoints (app.coolify.io or self-hosted instances).
[PROMPT_INJECTION]: The skill features commands to read application logs and manage environment variables. This represents an indirect prompt injection surface where malicious data within logs or variables could influence the agent's subsequent behavior.
Ingestion points: Reading application logs via applications logs and environment variables via applications envs list.
Boundary markers: None identified in the provided documentation to distinguish between data and instructions.
Capability inventory: File reading, network requests via curl, and local script execution.
Sanitization: No explicit sanitization or filtering of log content or variable values is described.

coolify