Agent Skills for Coding Work

Start with the current uncertainty. Debugging investigates why behavior is wrong; TDD makes an expected behavior executable; refactoring improves structure while preserving behavior. A single task may move through all three, but they answer different questions.

Provide the smallest reproducible example, expected and actual behavior, relevant errors, and the environment where the issue occurs. A failing test or concrete input is more actionable than a general statement that the code is broken.

A test is valuable when it can detect the behavior or regression at issue. Exploratory work may first need a clearer requirement; cosmetic changes may need visual checks. Avoid tests that merely repeat the implementation without checking a meaningful outcome.

Record the behavior that must remain stable, run relevant checks, and review the diff for accidental scope changes. Passing tests are useful evidence, but untested interfaces and error paths may still need targeted inspection.

Use it when the problem involves React or Next.js concerns such as rendering, data fetching, state, or bundle size. Measure the actual application so a plausible optimization does not become an unverified change with no demonstrated benefit.

Yes, when their roles are clear. For example, reproduce a failure, isolate the cause, make a small change, and then inspect the diff. Resolve conflicting instructions and keep one acceptance criterion rather than layering unrelated checklists.