@socai-io/dsh-socai
DeepSeek Harness 的 socai：由本地 socai CLI 支持的类型化小红书研究工具
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:socai-io/dsh-socai#438865ec35c97a03bb0cee50060c0123a6c5160eSocai 以 @socai-io/dsh-socai 发布，当前版本为 0.1.1。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
dsh-socai — Research, right inside your agent
Xiaohongshu research tools for DeepSeek Harness, powered by your local SocAI runtime.
dsh-socai adds three typed, read-first research tools and one guided Skill to
DeepSeek Harness. It uses the existing socai CLI for browser work, so your
login state, Chrome profile, and SocAI configuration stay in one place.
Search, author research, and note reading with dsh-socai
| Tool | Use it for |
|---|---|
socai_xhs_search | Preview Xiaohongshu results or run a focused deep search with filters. |
socai_xhs_author | Research a creator or competitor profile and scan their notes. |
socai_xhs_get_notes | Deep-read selected notes returned by search or author discovery. |
The bundled socai-xhs-research Skill guides the agent through a practical
preview → select → deep-read workflow. Model routing guidance teaches DSH when
to prefer these tools for Xiaohongshu research.
DeepSeek Harness
└─ dsh-socai tools + Skill
└─ local socai CLI (JSON)
└─ SocAI browser daemon
└─ your usable Chrome profile
The plugin launches socai directly with an argument array—never through a
shell. Calls are serialized because DSH sessions can share the same SocAI
browser daemon and Chrome profile. Results stay structured, model-visible output
is bounded, and local run paths are replaced with an opaque run identifier.
0.1.1-rc.2 or 0.1.2-alpha.2.0.4.12 or newer, available on the PATH inherited by DSH.Install SocAI on macOS:
curl -fsSL https://github.com/socai-io/socai/releases/latest/download/install.sh | sh
For other platforms and configuration options, see the SocAI repository.
Install the latest source from GitHub:
dsh plugin --profile web add git+https://github.com/socai-io/dsh-socai.git
Or pin the precompiled v0.1.1 release artifact:
dsh plugin --profile web add https://github.com/socai-io/dsh-socai/releases/download/v0.1.1/dsh-socai.tgz
Restart dsh web after adding or updating the bundle.
Ask DSH naturally; the model routing hint and Skill handle tool selection. For example:
Search Xiaohongshu for recent discussions about AI note-taking. Preview 20
results first, then deep-read the five most relevant notes.
Research this Xiaohongshu creator, summarize their positioning and recurring
topics, then compare engagement patterns across their latest 10 notes.
Open these selected notes from the previous search and extract the claims,
evidence, audience questions, and reusable content angles.
The tools are read-first. They do not publish, like, comment, follow, or bypass platform login and verification controls.
Override the dsh-socai row in the profile's cordis.patch.yml when needed:
- id: dsh-socai
name: '@socai-io/dsh-socai'
config:
binaryPath: /absolute/path/to/socai
timeoutMs: 900000
maxOutputBytes: 64000
productUrl: https://socai.io/?utm_source=deepseek-harness&utm_medium=plugin&utm_campaign=dsh-socai
| Field | Default | Purpose |
|---|---|---|
binaryPath | socai | Use an absolute CLI path when DSH does not inherit your shell PATH. |
timeoutMs | 900000 | Maximum duration for one SocAI browser operation. |
maxOutputBytes | 64000 | Maximum model-visible result size. |
productUrl | Tagged socai.io URL | Product/help link returned with useful attribution. |
CI builds, strictly typechecks, packages, and tests the plugin against both supported DSH revisions. The test suite covers:
Git installation, release-tarball installation, profile composition, and a live
dsh web boot are also verified as part of the release readiness record.
Use a built DeepSeek Harness source checkout:
DSH_CHECKOUT=/path/to/deepseek-harness node scripts/test.mjs
The build and test launchers are cross-platform Node scripts. The executable fixture integration suite runs on macOS/Linux; a Windows process-tree regression test is present, while a full Windows DSH + SocAI profile smoke test remains on the rollout checklist.
See FEASIBILITY.md for the architecture decision, current
compatibility evidence, risks, marketplace status, and go-live checklist.