luoyuctl/agenttrace — reverse-engineered prompt

Build me a local terminal tool that helps me understand what my AI coding agents have been doing and spending.

I want it to read the session logs already on my computer from tools like Claude Code, Codex CLI, Gemini CLI, Qwen Code, Cline, Aider, Cursor exports, OpenCode, Kimi CLI, Copilot style logs, and generic JSON or JSONL traces. When I open it, show a clean terminal dashboard with total sessions, tokens, estimated cost, time spent, health, slow runs, tool failures, and the worst sessions to inspect first.

Let me drill into one session to see why it was slow, where there were long gaps, retries, hanging tool calls, big inputs, errors, or context pressure. Add search and sorting by cost, duration, model, source, health, failures, and keywords.

Also make commands for doctor checks, JSON output, Markdown output, and a self contained HTML report. Keep everything private and local. Add tests, install scripts, and simple docs. Look up current docs online if needed.