Vendor replays
Coding agent risks with Postgres: replays and fixes
3 replays involve Postgres: 1 reported incident and 2 command classes. Each one shows the call, a fix that needs no extra tool, and what KYA did in the replay.
Agent wipes a production database during a code freeze
Fix: Give the agent a database role that can only read production and does not own the tables, and keep the development and production connection strings apart, so a development session never holds a production login. Turn on automatic backups with point-in-time recovery and test a restore.
Replay result: Stopped.
DELETE FROM a table with no WHERE clause
Fix: Give the agent a role that has no DELETE privilege on production tables and does not own them, and keep point-in-time recovery on. When a person runs a hand-written delete, start with BEGIN, check the row count, and run COMMIT only when it is right.
Replay result: Held.
TRUNCATE TABLE with CASCADE
Fix: Do not grant the TRUNCATE privilege to the agent's role, and do not make that role the owner of the tables, because owners can truncate, and CASCADE reaches every table that has a foreign key to the target. Keep point-in-time recovery on and test a restore.
Replay result: Stopped.