Pre-Mortem runs an adversarial critic against a planned change before deploy. Given the diff and the affected services, the critic enumerates the top three failure modes, ranks them by likelihood and severity, and proposes mitigations. The output is one paragraph per failure. Useful for changes that scare you (and changes that should but do not).
Pre-Mortem takes the diff (or the runbook), the affected services, the recent incident history, and the agent's own service knowledge. It runs an adversarial role: "what could go wrong?" The output is the top 3 failure modes, each with a one-paragraph explanation, a likelihood class, a severity class, and a one-line mitigation suggestion.
Configure Pre-Mortem as a CI gate: every PR that touches a tier-0 service runs it before merge. The author sees the failure modes in the PR comment and can either address them or note why they are accepting the risk. The gate is optional for lower-tier services.
For each failure mode, the author marks it as addressed (with a link to the mitigation in the PR), accepted (with a written justification), or dismissed (false-positive, with a reason). The decisions are stored with the deploy and surfaced in postmortems if the deploy turns out to cause an incident.
If a deploy turns out to cause an incident and the pre-mortem predicted the failure mode, the postmortem builder includes the prediction directly. "We were warned." This is not punitive; it is feedback for the gate. Patterns where pre-mortem predicts correctly raise the gate's influence; patterns where it consistently misses get prompt-tuned.
Subscribe to Nova AI Ops on YouTube for demos, tutorials, and feature deep-dives.
Pre-Mortem is the cheapest insurance against shipping a change that breaks production at 3am.