agent-usagelisted

# Agent Usage Analyze the shared usage archive with dot and DuckDB. Preserve the difference between recorded token usage, estimated cost, and the provider's actual bill. ## Workflow 1. **Bound the question**: harnesses, time range, sessions, models, and the comparison needed; inspect the existing archive before collecting more data. 1. **Use the owner**: dot owns source synchronization and archive schemas; read [queries.md](references/queries.md) for layout, fields, DuckDB queries, and exports. 1. **Analyze comparable data**: account for missing sessions, model aliases, cache tokens, provider accounting differences, and time zones before aggregating. 1. **Report**: include period, sources, completeness, units, assumptions, and the query or artifact supporting the result; protect prompt and account data. Use `dot agent stats --since 7d --by-model` for a combined prompt and token report, or add `--json` for analysis. API equivalents use the offline rate card in `agent.pricing`, independently of recorded cost; check `priced_sessions`, `pricing_complete` and `unpriced_reasons`. The default estimate assumes standard short context and 5-minute cache writes. Unknown models and unsupported accounting remain unpriced. Cursor session capture is not currently supported. ## Gotchas - **Unknown is not free**: Claude can report cost through `cost-state`; absent prices are `null`/`unknown`, not zero. Read `cost_known_sessions` and `cost_complete` before comparing cost. A known zero is