dba-rds-slow-query

Query pg_stat_statements ordered by mean_exec_time descending to find the slowest queries. This Skill flags any query with mean execution time above 1000ms and cross-references it with table scan statistics to identify the cause.

Query pg_stat_user_tables and calculate the sequential scan percentage. Tables with seq_scan_pct above 80% and more than 100 sequential scans are index candidates, which you cross-reference with slow query text to propose CREATE INDEX CONCURRENTLY statements.

You need read-only AWS permissions: rds:DescribeDBInstances, rds:DescribeDBParameters, and cloudwatch:GetMetricStatistics. On the database side, the PostgreSQL user needs the pg_read_all_stats role to read pg_stat_statements.

High CPU without individual slow queries usually indicates high query volume or connection saturation. Check DatabaseConnections in CloudWatch; if it exceeds roughly 130 on a db.t3.medium, connection pooling with PgBouncer is the recommended fix.

No, the Skill is strictly read-only and never executes DDL, parameter changes, or connection termination. It formulates recommendations such as index creation or work_mem tuning, which require DBA approval through a change window.

Do not use it for connection pool configuration issues, replication lag, or storage capacity problems, as those require separate diagnostic workflows. It is scoped specifically to query performance, index gaps, and parameter group review.