wxxb789/dsh-raven-research

README 摘要

dsh-raven-research One progressive, source-grounded Task for deep research, writing, and learning — inside DeepSeek Harness. Early checkpoints you can steer mid-run · citations verified against the bytes actually retrieved · no second agent runtime. English · 中文 TL;DR · Install · Usage · How it works · Configuration · FAQ [!IMPORTANT] v1 developer preview. Pinned and tested against DeepSeek Harness 0.1.1-rc.1 , which is itself an RC and ships breaking changes. Not published to npm yet — install from a checkout. TL;DR - What it is: a DeepSeek Harness ( dsh ) plugin that adds one progressive, evidence-aware Task abstraction for deep research, general writing, academic writing, and learning. - Why it matters: you get a useful Checkpoint early, you steer it mid-run instead of restarting, and every citation is checked against the bytes actually retrieved — not against what the model remembers. - How it is built: one Cordis plugin on the host plane, one model-facing raven task tool, one compact prompt section, and one settings card in the Web GUI. No second agent runtime, no model host, no vector store, no database. The Harness agent keeps researching and writing with its normal tools. -…

分类