The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/backup_pg.py utilizes subprocess.run to invoke system binaries such as pg_dump, psql, pg_restore, and pg_ctl. These calls are used to perform legitimate database operations like creating dumps, restoring from backups, and promoting replicas.
[INDIRECT_PROMPT_INJECTION]: The skill features a surface for indirect prompt injection as it ingests user-provided database names, hostnames, and credentials. These inputs are used to construct shell commands and SQL statements (e.g., CREATE DATABASE in scripts/backup_pg.py). While there is no explicit sanitization for these inputs in the provided script, this is an inherent characteristic of database administration tools designed to be operated by a trusted agent.
[SAFE]: The skill's behavior is consistent with its stated purpose of PostgreSQL administration. It does not exhibit signs of data exfiltration, obfuscation, or persistence outside of standard database features (e.g., pg_cron). The management of sensitive credentials via environment variables in the backup script follows standard security practices.

postgres-pro