What problem does it solve? When a bug writes wrong values into the production database, fixing the code alone is not enough because the bad data is already persisted. This Skill provides a disciplined playbook for diagnosing the damage, writing a narrowly scoped one-time heal, and verifying it without touching production directly. ## Core Features & Use Cases - Structured heal plan: Requires a written diagnosis, affected-row scope, deterministic repair rule, idempotency strategy, rollback posture, and verification criteria before any code is written. - One-time idempotent heal pattern: Implements heals in artifacts/api-server/src/lib/dataHeals.ts using a single transaction, a claimed marker row in data_heals, and monotonic updatedAtMs bumps so stale clients cannot republish poisoned values. - Manager-value preservation: Excludes or escalates rows that may be legitimate manager corrections, and restricts daily_sync repairs to today and future dates. - Use Case: An import bug wrote a wrong number into many brand profiles. Use this Skill to survey prevalence with grouped counts, dry-run the target set, ship a marker-claimed heal transaction, and verify the repair in dev and after publish. ## Quick Start Ask the AI to use the data-heal-playbook to diagnose and repair the poisoned profile values left behind by the fixed import bug.
