observability

Review or write a database schema/data migration so it ships without downtime: expand-contract sequencing, lock analysis, safe index creation, and batched…

Drive a production incident: triage severity, mitigate first, then diagnose, communicate, and write a blameless postmortem with real action items.

Summarise the current session into a handoff document so a fresh agent window can continue without losing context.

Review a change to an API contract for compatibility before it ships, across REST/JSON, gRPC/protobuf, or Thrift IDL.

Retire an API, endpoint, service, table, or feature without breaking the people still using it: usage measurement, advisory vs.

Write documentation that stays true, and find what a change just made false.

Prepare work for review without ever publishing it: branch hygiene, deliberate staging, atomic commits, short commit messages carrying no agent attribution…

Turn agreed work into a written plan file with verifiable steps, then execute it one step at a time with the file as the source of truth.

Build or review a UI so it works with a keyboard, a screen reader, and low vision: semantic HTML, focus management, labelled forms, accessible errors…

Reduce working code to its simplest correct form without changing behavior.

Wire a UI to a backend API correctly: request waterfalls, race conditions, the four UI states, caching server state, optimistic updates with rollback…

Write or review React components against the patterns that actually cause bugs: unnecessary effects, derived state, stale closures, key misuse, re-render…