copilot-session-costlisted

# Copilot session cost report This skill runs one read-only script, `scripts/session-cost-report.py`. The script reads local Copilot CLI storage only: - `~/.copilot/session-store.db` (table `assistant_usage_events`) - `~/.copilot/session-state/<id>/` (used only to find the active session) It reports call counts, token counts (input, output, cache-read, cache-write, reasoning), a per-model breakdown, and root-vs-subagent totals, in AI credits. It also reports **cost concentration**: per primary metric (token type, calls, AI credits), which model, agent, initiator, or time window holds an outsized share of that metric. ## Overall token mix Before the concentration drilldown, the report shows a concise **overall token mix** that still gives all 5 recorded counters (input, output, cache-read, cache-write, reasoning) a visible percentage or ratio — but never claims they form one partition: - **Top-level**: `input_tokens` and `output_tokens` as a percentage of their own sum. This is the one safe, non-overlapping pair — neither field can include the other. - **Cache ratios**: `cache_read_tokens` and `cache_write_tokens`, EACH shown as its own independent ratio to `input_tokens` — shown only when the schema has real cache columns and `input_tokens > 0`. The two ratios are not claimed to sum to 100%. - **Reasoning ratio**: `reasoning_tokens` shown as its own ratio to `output_tokens` — shown only when the schema has a real `reasoning_tokens` column and `output_token