The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill provides templates for full administrative control over production databases.
Evidence: Use of pg_dump, pg_restore, and psql for arbitrary command execution.
Evidence: Emergency procedures include pg_terminate_backend(pid) which can be used for Denial of Service (DoS) attacks.
[DATA_EXFILTRATION] (HIGH): High capability for bulk data movement.
Evidence: pg_dump -d production > backup.dump facilitates extracting the entire database content to the local filesystem.
Evidence: pg_basebackup -h primary_host enables streaming entire database clusters across the network.
[CREDENTIALS_UNSAFE] (HIGH): Contains hardcoded passwords in usage examples.
Evidence: CREATE USER replicator WITH REPLICATION ENCRYPTED PASSWORD 'secret'; in SKILL.md.
Evidence: CREATE ROLE app_user WITH LOGIN PASSWORD 'secure_password'; in SKILL.md.
[INDIRECT PROMPT INJECTION] (HIGH): The skill processes untrusted external data with high-privilege capabilities.
Ingestion points: Reads from pg_stat_activity, pg_stat_statements, and pg_tables which can contain attacker-influenced strings (e.g., query comments, table names).
Boundary markers: None identified; untrusted data is directly interpolated into agent reasoning for 'Health Checks' and 'Troubleshooting'.
Capability inventory: Shell execution (pg_dump), system-level SQL execution, and network connectivity.
Sanitization: Absent. Malicious SQL queries or table names could influence agent logic during maintenance tasks.

managing-databases