The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Fetches and executes environment installation scripts from trusted organizations and well-known services, including the official Docker installer, NodeSource setup scripts for Node.js, and the Rustup installer for Rust.
[EXTERNAL_DOWNLOADS]: Downloads Go language binary archives directly from the official go.dev domain during the environment setup phase.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by extracting data from local project files and using it in shell commands.
Ingestion points: Metadata and configuration values are read from package.json, .env, main.py, app.py, manage.py, Cargo.toml, and go.mod in the project root.
Boundary markers: No specific delimiters or instructions are used to distinguish untrusted file content from system commands.
Capability inventory: The agent has access to the Bash tool to execute local commands and remote commands via ssh and rsync with root privileges.
Sanitization: Extracted variables like $PORT, $PROJECT_NAME, and $DOMAIN are interpolated directly into shell strings without validation or escaping.
[COMMAND_EXECUTION]: Performs administrative tasks on remote servers via SSH, including managing system services via systemctl, installing packages with apt-get, and modifying critical configuration files in /etc/ssh/ and /etc/nginx/.
[CREDENTIALS_UNSAFE]: References the use of the sshpass utility to handle passwords during initial server configuration when SSH keys are not yet established.
[DATA_EXFILTRATION]: Accesses sensitive files such as .env configuration files and SQLite databases during the deployment process to identify project settings or adjust file permissions.

server-deploy