A new open-source plugin called Goyangi is addressing the opacity of agentic coding workflows by providing a real-time panel alongside Claude Code. The tool, whose name translates to "cat" in Korean, sits next to the conversation to monitor every turn, breaking down where time and API costs are being spent. This release targets a critical pain point for developers using large language models for coding: the lack of granular visibility into context window consumption and token economics during autonomous sessions.

Granular Turn-by-Turn Analytics

Goyangi provides a detailed view of the turn flow, listing every model request, tool call, and subagent execution in chronological order. Each entry includes precise token counts, cost estimates, and a time-share bar to visualize latency contributions. The plugin also features local gauges for context window usage, cache hit rates, and the 5-hour and 7-day usage limits, offering a forecast of when limits might be reached. Crucially, all this analysis is computed locally, ensuring that monitoring the agent does not itself consume additional API tokens.

Session Management and Handoff Protocols

The plugin introduces a robust system for managing multiple concurrent sessions and browser tabs, identified by unique numbers. Users can issue commands like "send this to session 2" to pass a handoff note between Claude Code instances or "pull from 1" to retrieve context from another session. This functionality extends to interoperability with other coding agents such as Codex and OpenCode. When a session's context approaches its limit, Goyangi alerts the user, allowing them to click "Continue" to spawn a fresh session that inherits the work via a generated handoff note.

Visual Feedback and Terminal Integration

Distinct from standard log viewers, Goyangi includes a "walk view" where a pixel-art cat or dog animates through the turn. The character's mood visually represents the current state of context usage, cost, and limit proximity. This visual layer relies on Unicode 16 octant characters for high-fidelity rendering in terminals like Ghostty, kitty, and WezTerm. For terminals like iTerm2 or VS Code, the plugin falls back to a coarser quadrant-block drawing to maintain layout integrity. The panel is accessible via the /goyangi command and automatically opens at the start of new sessions.

Key Takeaways

  • Goyangi is an MIT-licensed plugin for Claude Code that visualizes token usage, costs, and cache hits per turn without consuming extra API tokens.
  • It facilitates cross-session and cross-agent handoffs, allowing users to manage context limits by spawning new sessions with preserved state.
  • The tool requires Claude Code with mod function hooks support and offers distinct visual fidelity based on the user's terminal capabilities.

The Bottom Line

As agentic coding tools become more autonomous, black-box execution is no longer acceptable; Goyangi proves that local, token-free observability is essential for managing the economics and context limits of modern LLM workflows.