Debug this code (with reasoning)When something breaks and you want a thorough diagnosis, not just a fix.
Refactor for readability (without changing behavior)You have working code that's hard to read. You want it cleaner without breaking anything.
Write tests for this function (with edge cases I haven't thought of)You have a function and need test coverage. The boring tests are easy. The edge cases are what catch real bugs.
Generate CRUD boilerplate (then justify the choices)You need basic CRUD endpoints/handlers and don't want to type it all out, but you also don't want generic AI slop.
Convert this code from X to YYou're migrating between languages or frameworks and need a translation that respects idioms of the target.
Add types to legacy JS / Python (without rewriting it)You have untyped code that works. You want types added incrementally without breaking it.
Architecture decision (build vs buy / framework choice)You need to make a non-trivial architecture call and want a structured pros/cons, not a vibe.
Three-level code review (quick / thorough / architectural)You need feedback on a PR but you also want context-appropriate depth, not a single review style.
Generate a database migration scriptYou're changing schema and need a migration that's safe to run, not just syntactically correct.
Verify AI-generated code before you commitYou wrote code with AI help. Before committing, run a verification pass that checks the parts AI is most likely to invent.
Escape an AI agent loopYour agent has tried 3+ times and the test still fails. Diagnose what's actually wrong and produce a fresh-session restart prompt.
Build a context bundle before asking AI to codeFor non-trivial coding tasks: prepare the context bundle (real schema, types, signatures) that prevents hallucinations before they happen.
Design a verification gate for an AI agent stepAn agent step produces output (code, data, an action) and you want to define what gets checked before that output is trusted downstream.
Review AI-generated output as a hypothesis (not a deliverable)An AI agent produced an output (code, doc, plan). Interrogate it for what was actually verified vs assumed vs invented, before you trust it.
Audit a spec for hidden assumptions before handing it to an agentYou have a spec, build plan, or design doc and want to find the gaps before an AI agent or junior engineer takes everything literally and fills them with guesses.
Scope a new build phase with testable acceptance criteriaA vague "next thing to build" needs to become a phase-gated chunk of work with entry conditions, scope, and a verifiable exit gate.
Generate a debugging hypothesis tree (ranked by likelihood × cost)Something is broken. You want a structured set of hypotheses with cheap verification steps, not a stack-overflow random walk.
Design a Postgres data model for a featureAdding a feature that needs new tables/columns, or sketching a new system from scratch. You want a model that's correct first, fast second, clever never.
Write a technical decision record (ADR) for future-youAn architecture choice you'll forget the reasoning for in 3 months (framework, runtime, deployment target, package structure). Make it legible and revisitable.