ccglass - See what your coding agent sends to the model

I found ccglass while debugging OpenAI-compatible coding-agent traffic. It gives you a local dashboard for the parts that are usually hidden: actual request bodies, system prompts, tool schemas, tool calls, tool results, streamed responses, token usage, cost, latency, and diffs between turns. Useful when an agent calls the wrong tool, sends too much context, breaks on streaming tool calls, or behaves differently across providers. It supports Claude Code, Codex, OpenCode, Qoder, Ollama, LM Studio, OpenRouter, Reasonix, DeepSeek-TUI, and other OpenAI-compatible setups. GitHub: https://github.com/jianshuo/ccglass