dsh-wisp-science-lab

English | 中文

A DeepSeek Harness plugin that turns your agent into a laboratory PI for local Wisp Science projects.

dsh advises. Wisp runs the experiments. v0 is read-only — list projects, take a bounded progress snapshot, give evidence-backed next steps. It does not write wisp.sqlite, does not run Python/R/FASTQ inside dsh, and cannot dispatch the Wisp agent yet.

You  →  dsh (PI)  →  wisp_list_projects / wisp_project_snapshot
↓  copy sqlite + wal + shm, read-only
Wisp Science desktop (experimenter)

Why

• See every local project — names, session/run/artifact counts, Windows and WSL paths.
• Bounded snapshots — recent runs, artifacts, research-graph titles, .wisp/WISP.md, memory filenames. Caps so a project with 358 runs does not flood the context.
• PI persona — must cite Run titles / artifact names / WISP.md. Forbidden to invent analyses that are not in the snapshot.
• Safe on a live desktop library — copies wisp.sqlite (and -wal/-shm if present) to a temp dir, never checkpoints the original.

Install

This plugin runs inside dsh. Install DeepSeek Harness first.

1. DeepSeek Harness

Requires Node.js ^22.19.0 or >=24.

npx @deepseek-ai/dsh web

Open http://127.0.0.1:3080 and add a model API key in Settings.

To get a global dsh command:

npm install -g @deepseek-ai/dsh
dsh web

Source checkout: deepseek-harness README.

2. This plugin

dsh plugin --profile web add github:xuzhougeng/dsh-wisp-science-lab
# if dsh is not on PATH:
npx @deepseek-ai/dsh plugin --profile web add github:xuzhougeng/dsh-wisp-science-lab

That both installs the package and registers cordis.patch.yml on the profile (dsh.bundle). Do not --patch the same layer on top — it will load twice.

Restart:

dsh web

You should see [wisp-science-lab] plugin loaded.

From a local clone:

git clone https://github.com/xuzhougeng/dsh-wisp-science-lab.git
dsh plugin --profile web add ./dsh-wisp-science-lab

Not published to npm yet — do not npm i dsh-wisp-science-lab.

dsh plugin --profile web remove dsh-wisp-science-lab

Usage

Talk to the agent in dsh:

• List my Wisp projects
• How far is “转录组分析”? What should we do next?
• Compare progress of project A and project B

The model should call wisp_list_projects / wisp_project_snapshot instead of bash-scanning the disk. Advice must quote snapshot evidence. If you ask it to “go run the experiment”, v0 only returns a prompt you can paste into Wisp.

Config

The plugin reads Wisp’s global wisp.sqlite. Resolution order:

1. appDataDir in plugin config (if non-empty)
2. WISP_APP_DATA_DIR
3. Windows: %APPDATA%\science.wisp-science\wisp-science
4. Linux: ~/.local/share/science.wisp-science/wisp-science

If dsh runs in WSL and the library lives on Windows, set (use your Windows username):

export WISP_APP_DATA_DIR=/mnt/c/Users//AppData/Roaming/science.wisp-science/wisp-science

Or in ~/.dsh/profiles/web/cordis.patch.yml:

- id: wisp-science-lab
config:
appDataDir: /mnt/c/Users//AppData/Roaming/science.wisp-science/wisp-science

Limits

• Read-only. Does not write the live library or project trees.
• WAL copy. Direct sqlite3 on a locked Windows file from WSL often hits disk I/O error. Each tool call copies sqlite + wal + shm, then deletes the temp dir.
• Paths. Store rows are usually Windows paths. On WSL the result also includes workspaceDirLocal (D:\foo → /mnt/d/foo when that path exists).
• No full transcripts. Snapshots never dump complete messages.content.

Develop

Source overlay (do not stack on a marketplace install):

cd /path/to/deepseek-harness
pnpm dsh web --patch /path/to/dsh-wisp-science-lab/cordis.dev.yml

Point cordis.dev.yml at your checkout. After editing src/, run pnpm build and commit lib/index.js.

pnpm test
pnpm print-snapshot            # live copy, else testdata/mini.sqlite
WISP_LAB_LIVE=1 pnpm test      # optional; skips on lock

This repo is tagged dsh-plugin so dsh-find-plugin and plugin markets can index it.