Jay E’s RoboNuggets 'ribbon' mod for Claude Code has long provided users with a visual band displaying weekly and 5-hour usage limits alongside prompt cache status. However, a recent fork by developer Kompas has emerged to resolve persistent usability issues, most notably the failure to update usage data when switching between Claude accounts. The original mod retained stale data from previous sessions, a flaw that became particularly confusing for users toggling between local models and Anthropic’s hosted services.

Account Switching and Local Model Support

The fork introduces a robust account detection mechanism that reads organization and account IDs directly from Claude Code’s configuration files. When a user switches accounts, the band immediately clears old metrics and fetches fresh data via the /usage command endpoint, ensuring that limits reflect the current session before any prompts are sent. For local models, which lack plan-based limits, the mod now correctly identifies the model name and context window fill status, removing misleading cache indicators that previously suggested Anthropic’s caching behavior applied to local inference.

Correcting Pricing Table Discrepancies

Beyond UI improvements, Kompas conducted a rigorous audit of the mod’s internal pricing table against Anthropic’s official rates, discovering three significant errors. The original ribbon overestimated the cost of warm cache reads for Sonnet 5.5 by 100%, listing it at $0.20 per million tokens instead of the correct $0.10. Haiku 5.5 prices were inflated by 5x for short prompts, and Fable 5.2 rates were incorrectly set at $10/$0.25 rather than the accurate $8/$0.20. These corrections are critical for developers relying on the band for real-time cost estimation during complex coding sessions.

Customization and Visual Refinements

The fork addresses aesthetic complaints by aligning the band’s color scheme with the user’s active Claude Code theme, replacing the original’s jarring orange with Claude’s native coral. Users can now toggle a 'minimal' style that hides dot bars and displays only numerical percentages, or enable 'colors muted' for a grayscale interface that only highlights warnings in color. The labels have also been clarified, changing cryptic abbreviations like '7D' to explicit terms like 'Week' and '5-hour', while a new /ribbon command prints a legend to explain cache status and reset timers.

Key Takeaways

  • The Kompas fork fixes critical bugs where usage data failed to update upon switching Claude accounts.
  • Three major pricing errors in the original ribbon were corrected, impacting cost estimates for Sonnet, Haiku, and Fable models.
  • New configuration options allow for minimal visual modes and theme-aligned colors, addressing user complaints about the original design.
  • Local model support has been improved to accurately display context usage without misleading cache information.

The Bottom Line

While the original ribbon was a clever hack, this fork transforms it into a reliable tool for serious developers who cannot afford stale data or incorrect cost projections in their workflow.