@yfzhou/dsh-session-cost-statusline

Adds an estimated session cost in CNY before the standard DeepSeek Harness composer statistics.

The plugin registers a durable sessionCost projection on the host. It folds each logged request step and prices each assistant/message.usage record using the model from its request/header. The client only renders that projection, so totals replay for existing sessions and remain correct when a session changes models.

It currently recognizes selected DeepSeek V4 and GPT-5.6 models. Usage from an unknown model is deliberately omitted rather than priced with an unrelated fallback, so the displayed figure is a lower-bound operational estimate rather than billing authority. DeepSeek model ids that carry a deployment build tag (for example the Ark/ByteDance id deepseek-v4-flash-ga-260731) are normalized to their DeepSeek V4 family before pricing, so any build of a known family is priced.

Install

dsh plugin --profile web add @yfzhou/dsh-session-cost-statusline

The package supplies its own DSH bundle row; do not add a separate cordis.patch.yml entry.

Runtime requirements

• Host service: sessionProjections
• Client service: slots
• Client slot: conversation.composer.dock
• Tested with DSH 0.1.0-rc.6

DeepSeek V4 uses the published effective-dated schedule from the DeepSeek pricing documentation: starting 2026-08-17 00:00 Beijing time, peak hours are 09:00–12:00 and 14:00–18:00, with all other hours off-peak. Per million tokens, Flash is ¥1.50/¥0.05/¥4.50 off-peak and ¥3.00/¥0.10/¥9.00 peak; Pro is ¥4.50/¥0.15/¥13.50 off-peak and ¥9.00/¥0.30/¥27.00 peak (uncached input/cache-hit input/output). Sessions before that effective time use the then-published fixed prices: Flash ¥1/¥0.02/¥2 and Pro ¥3/¥0.025/¥6.

The fold prices each model request separately: model switches are attached to the step that uses the header, and every step selects peak/off-peak from its own persisted request start (request/header, or step/start when the header is unchanged), using Beijing time explicitly. It does not split a request that crosses a pricing boundary; all of that request's tokens use the bucket at request start. Pricing changes over time; review the embedded rate table before relying on the displayed amount.

License

MIT