The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions (e.g., in examples/03-performance-tuning.md) guide the agent to analyze execution plans and database statistics using tools like EXPLAIN ANALYZE and pg_stat_statements. These outputs include query text and metadata that may be influenced by untrusted database users, serving as an ingestion surface for indirect prompt injection.
Ingestion points: Query statistics from pg_stat_statements, execution plans from EXPLAIN, and system logs.
Boundary markers: None specified in the skill instructions to delimit external data.
Capability inventory: The skill assumes the agent can perform SQL execution and administrative shell operations (systemctl, pg_basebackup, pg_ctl).
Sanitization: No explicit sanitization of database outputs is mentioned.
[COMMAND_EXECUTION]: The skill contains documentation for high-privilege administrative commands, including systemctl restart postgresql, pg_basebackup, and file system operations like rm -rf /var/lib/pgsql/data/* in examples/04-streaming-replication.md. While these are standard for database replication guides, they involve powerful system-level actions.
[CREDENTIALS_UNSAFE]: The documentation includes example credentials like Str0ng!Pass and strong_password in examples/04-streaming-replication.md and references/08-replication-backup.md. These are used as placeholders within setup tutorials and do not represent leaked production secrets.

postgresql