I wrote a profiling tool for agents

While building an agent for a client, we hit a scaling problem: a coordinator agent had 4-5 subagents each with 15-20 tool calls. Each tool result got appended to the context, then what started as a r