What problem does it solve? Long-running LLM tasks consume large volumes of tokens, but raw session logs do not directly show how much money was spent. This Skill converts Codex/OpenAI-style token usage into a reproducible RMB-Cost.md report that breaks down uncached input, cached input, and output costs in both USD and RMB. ## Core Features & Use Cases - Session-based usage extraction: Parses Codex session JSONL token_count events within a start/end timestamp window to compute delta token usage. - Dual-model cost comparison: Calculates costs for both GPT and DeepSeek pricing tiers, separating cache-hit, cache-miss, and output components. - Manual mode and price overrides: Accepts manually supplied token counts and custom per-million-token prices plus USD/CNY exchange rates. - Use Case: After a 3-hour automated experiment finishes, run the script against the session JSONL to produce RMB-Cost.md showing 320M input tokens, cache hit savings, and total RMB cost under both GPT and DeepSeek pricing. ## Quick Start Ask the assistant to use the rmb-cost-report skill to generate an RMB-Cost.md report from your Codex session JSONL file with the goal's start and end timestamps.
