A recent experience with multi-agent code reviews in Claude Code's 'ultracode' mode shed light on the rapid consumption of token quotas. The first review involved 31 agents and the second used 19, with each agent consuming between 50,000 and 80,000 tokens. Despite the substantial volume of output, much of it was unusable, and the cost was only realized post-facto.
To regain control over quota visibility, the /usage command was employed, followed by a persistent status line to monitor token usage in real-time. Claude Code's multi-agent approach to code reviews distributes the workload across multiple agents, each tackling different aspects of the code. For instance, a review of gapless HLS playback (#1458) involved 31 agents and five phases of analysis, while the ruby_llm migration (#1504) used 19 agents, demonstrating the breadth of the technique.
The initial realization was that the cost incurred was not proportional to the output quality, and the lack of real-time cost visibility was a significant drawback. The /usage command provides an accurate overview of token usage, but it is a manual process that users often overlook. To address this, ccusage, a small open-source CLI tool, was introduced to continuously monitor and report on token consumption. However, ccusage operates based on local logs and pricing tables, thus cannot accurately display the percentage of the quota remaining.
The real-time quota limit is only accessible through the /usage command and the rate_limits block that Claude Code provides to status line scripts. A small script was developed to display this information as a bar at the bottom of the terminal, which was then improved using the ccstatusline tool, offering pre-built widgets for session and weekly usage with themes and reset timers.
Ultimately, this experience highlighted the importance of real-time monitoring tools to manage Claude Code quotas effectively. With the right tools in place, users can now better gauge their available token quota before initiating heavy reviews with multiple agents, ensuring more efficient utilization of resources.