If you're testing Claude skills with Agentica's async REPL harness, this prevents common async/await mistakes that silently break your code. It enforces two critical rules: always await tool calls that return futures (like view_file or ask_memory), and keep your logic in a single code block since the REPL only executes the first one. The second rule catches people off guard because splitting compute and return across blocks feels natural but means your return statement never runs. It's narrowly scoped to this specific testing environment, so you won't invoke it directly, but it'll save you from debugging why your tests are failing mysteriously when the issue is just missing await keywords or block splitting.
npx skills add https://github.com/parcadei/continuous-claude-v3 --skill async-repl-protocol