概览
dsh-chat-cost
源码级技术说明DeepSeek Harness (dsh) plugin: live token cost of every chat, its subagents and the whole session tree, the price of each finished answer in the transcript, budget planning that packs a work plan under a money limit, and scenario comparison that names which models are worth connecting — DeepSeek, OpenAI, Anthropic, Gemini, Kimi, Grok and Mistral prices from a bundled catalog, plus an append-only JSONL cost log in your project folder.收起技术说明
README / ZH
插件文档
目录摘要
dsh.pub 核对固定版本的组合包契约、运行时事实与分发语义；完整 README 请查看源仓库。
在 GitHub 阅读完整 READMELIMITATIONS
已知限制
- Reasoning tokens are billed as output by the providers and are counted as output here. - Long-context tier prices published for some Gemini and Grok models are not applied yet; the base tier is used. - A chat that the log has never seen and that is not open shows `—`: the plugin prices what it can prove (the live session, or the log) and says so rather than estimating from a projection it cannot read. - The live readout counts only the tail of the log (`ledgerTailBytes`, 2 MiB by default); older spend stays in the file and the summary reports the truncation. - Per-answer badges exist for turns the Host can price. A chat taken from the log has a total and no turn boundaries, so it shows no badges rather than invented ones. - The write queue serializes the plugin's own writes. A human editing `plan.json` at the moment the model writes it can still lose that edit; the file is small and meant to be reviewed, not co-edited.