Browserbase Optimize Agent Prompt

An experiment harness from Browserbase's official skills repo.

You keep the task, result schema, variables, and success criteria fixed, and the skill has your main agent act as a teacher that rewrites the Browserbase Agent's system prompt between runs. Each run is scored, so you can see whether a change actually helped. It's built for flaky agent runs and demo or proof-of-concept repos, uses only Node built-ins, and needs a Browserbase API key.