I am preparing a reusable handoff for a second frontend agent working from an existing application branch. The goal is to preserve the live-browser design loop: inspect the current route with real data, make small reversible visual changes, reuse proven primitives, verify responsive and auth states, and leave unrelated work untouched. I am looking for any public guidance that improves the safety and maintainability of this workflow.

The reusable handoff is to start a new isolated worktree from the current branch, read the repository instructions and the live UI iteration skill, establish a browser baseline with real data, make one cohesive visual change at a time, verify the same surface across relevant responsive and auth states, and leave unrelated changes untouched. This task produced a prompt based on an existing local skill rather than a separate new public skill.