Loading prompts
Versioned, tested prompts from teams that ship them. Every result is scored against multiple models.
Reviews a diff for correctness, resource handling and interface breakage. Ignores style.
Writes a blameless incident summary from a raw timeline, without inventing a cause.