Lessons from skilling up coding agents to use Langfuse

Claude Code attempts to add Langfuse instrumentation using stale pre-training context, producing broken traces. It then catches the failure and fetches current documentation to correct itself, highlighting the need for better agent context handling.