Preface¶

In the plugin-based development scenario of DSH, when a model needs to call external tools, it is usually considered first whether the tool dependencies are lightweight and whether the runtime requires network access or additional software. Blaczz/dsh-sci is designed for such needs, providing three directly registrable tools: physical unit conversion, CODATA physical constant lookup, and ODE/dynamical system simulation. It does not modify the DSH core or introduce runtime dependencies.

What This Is¶

Blaczz/dsh-sci is an independent community plugin maintained by Blaczz, licensed under MIT © 2026 Blaczz, and is not affiliated with DeepSeek Harness. It addresses the need for lightweight scientific computation tool calls within DSH: unit conversion, constant lookup, and fixed-model ODE simulation.

Core Features¶

Physical Unit Conversion¶

Provides conversion among more than 70 units, covering dimensions such as length, mass, time, temperature, energy, pressure, frequency, force, angle, volume, area, and electric.

CODATA Physical Constant Lookup¶

Supports querying CODATA physical constants, including c, h, e, k, G, N_A, and others.

ODE / Dynamical System Simulation¶

Uses the fourth-order Runge–Kutta method to perform ODE/dynamical system simulation on 9 preset models.

Installation and Enablement¶

First, ensure the environment meets the prerequisites: DSH is available, and dsh web can run; Node ≥ 22.19; pnpm.

Then, execute the following GitHub installation command:

dsh plugin --profile web add "github:Blaczz/dsh-sci#main"

After installation, restart dsh web to activate the plugin registration.

Typical Usage¶

Below are examples of the three types of tools as described in the plugin documentation.

Unit Conversion¶

Convert 1 eV to joules:

units_convert(value=1, from="eV", to="J") → 1.602176634e-19

Physical Constant Lookup¶

Return constants such as c, h, e, k, G, N_A:

physical_constants() → c, h, e, k, G, N_A, … all

Calling Preset ODE Models¶

Using the lotka-volterra model as an example:

ode_solve(model="lotka-volterra", params={...}, tEnd=50) → returns the time series + final state [prey, predator]

Implementation Features¶

• Zero dependencies: Pure function implementation, no runtime node_modules, no network access, and no reliance on external software.
• No core modifications: Tools are registered via ctx.tools.register without including a client bundle.
• Optional peer dependencies: @deepseek-ai/cordis >=4.0.1-rc.1 <5.0.0, @deepseek-ai/dsh-tools >=0.1.0-rc.3 <0.2.0.

Applicable Scenarios and Notes¶

Suitable for users who need to call physical unit conversion, CODATA constant lookup, and fixed ODE model simulation within DSH conversations or agent workflows. Since the plugin runs with the current dsh process permissions, the source code, dependency declarations, and license should be reviewed before installation; if the aforementioned optional peer dependencies are available in the environment, verify that the version ranges are compatible with the current DSH configuration.

Links¶

• Community directory: https://www.skillhub.cn/plugins/Blaczz/dsh-sci
• GitHub: https://github.com/Blaczz/dsh-sci