session-stats

Renders the quantitative stats of a Claude Code session as a single self-contained, dark-theme HTML page: KPI cards (duration, prompts, tool calls, files written/edited, output tokens, est. cost), a per-tool usage table with share bars, and a token-bucket table (input, output, cache created, cache read). Stats only - no prompts, observations, tone analysis, or recommendations. Two-stage offline pipeline with no external dependency: scan_jsonl.py parses a Claude Code .jsonl transcript and computes every metric, including a cost estimate derived from each assistant message's token usage times per-model pricing (current Opus 4.8 catalog rates, with IN_RATE/OUT_RATE/CW_RATE/CR_RATE env overrides); build_stats_html.py injects the result into a fixed dark template and writes a standalone document with no <script> tags or external assets. promptCount excludes slash-command and system turns. Cost method groups assistant messages by model and sums input/output/cache-write/cache-read tokens times their rates divided by one million. Includes a /session-stats slash command that resolves the current session transcript (or an explicit path) and runs the pipeline.