Choose a project you can run locally, such as a small to-do page. Add one status filter. Give the agent the repository and a short brief. Ask it to locate the entry point, inspect existing components and test commands, and identify the files it plans to change.
Define observable results
Prepare three tasks with one completed. Active should show two and All should restore three. Specify the empty state and default selection. Preserve add, complete, and delete behavior. Track persistence changes as a separate task.
Check the implementation as you go
Reuse existing controls. Use Frontend Design for the relationship between filters and results. Run the page locally, check the browser console, and try keyboard use and a narrow viewport. Use Webapp Testing to turn the fixture into a repeatable check.
Record what someone else needs
Keep the start command, changed files, fixture, actual results, and any unverified behavior. If a run fails, provide the first error and reproduction steps before asking the agent for a focused fix.