Introduction¶

The DSH plugin structure allows tools to be added to the profile’s layer stack. For agents, many decisions depend on the current environment: what the platform is, what the shell is, which commands are in PATH, what the software versions are, CPU, memory, and how much disk space is left. If these are guessed by the model, or tried one by one with multiple slow discovery commands, it is unstable and wastes tool calls.

Below is an introduction to @deepseek-ai/dsh-tool-scout. It provides the environment_probe tool, which reads the runtime environment, PATH commands, software versions, and host resources on demand, outputting a normalized JSON object while preserving human-readable text.

What is it¶

It is a model-visible environment probing tool for DeepSeek Harness, installed as a dsh bundle, comes with cordis.patch.yml, and joins the profile’s layer stack.

The package name is @deepseek-ai/dsh-tool-scout, version 0.0.1-rc.2, license MIT, package.json marked private: true. package.json marks author as ExDragine; repository path clue is omdsh-dev/dsh-scout, README title is dsh-tool-scout. The scraped data does not directly state the final maintainer.

Core Features¶

Provides the environment_probe tool.
scope supports environment, commands, software, resources, apps, serial, usb, network, gpu, ports, services, workspace, printers.
names can limit command, software, app, or printer names; invalid for environment, resources, serial, usb, network, gpu, ports, services, workspace.
Probes runtime environment, such as platform, arch, hostname, user, home, cwd, Node, shell, timezone, locale, endianness, PID, dsh home.
Detects commands resolvable on PATH.
Detects installed software and versions.
Detects host resources like CPU, memory, disk, etc.
Detects TCP listening ports and visible processes.
Detects system service status.
Detects workspace toolchain signals.
Detects hardware or host information like serial, usb, network, gpu.
apps is only available on macOS; other platforms report unavailable.
All scopes are read-only; does not invoke shell based on model input; only reads a curated environment variables allowlist, does not read the full process.env.
Version probing uses execFile(name, ['--version']), with timeout, concurrency limit, and TTL caching.
Outputs a normalized JSON object while preserving human-readable text.
The plugin contributes a tool:scout prompt section, suggesting that probing should be done before assuming platform, shell, command availability, software versions, resource limits, listening ports, service status, project toolchains, or printers, and only requesting the needed scopes.

Installation and Enablement¶

The package is installed as a dsh bundle. Verified documentation provides the installation command in the form of a local path:

# from the public Harness source checkout:
cd /path/to/deepseek-harness
pnpm dsh plugin --profile web add /path/to/dsh-scout

After execution, the dsh process needs to be restarted. dsh.profile.bundles in the profile will include @deepseek-ai/dsh-tool-scout, and the environment_probe tool will appear in the tool list.

Verified documentation does not provide a registry installation command; do not concatenate other installation forms based on the plugin name.

Typical Usage¶

The default scope is environment. scope can be a single category, an ordered array of categories, or all.

{ "scope": "environment" }

If you only care about hardware or host information related to robot debugging, the documentation example is:

{ "scope": ["serial", "usb", "network"] }

names can limit the command, software, app, or printer names being probed; however, it is invalid for environment, resources, serial, usb, network, gpu, ports, services, workspace.

apps is only supported on macOS; other platforms will report unavailable.

The prompt section contributed by the plugin suggests: calling environment_probe before assuming platform, shell, command availability, software versions, resource limits, listening ports, service status, project toolchains, or printers, and only requesting the needed scopes.

Configuration¶

Configuration is located in the profile patch layer (cordis.patch.yml), belonging to a Host-side-only out-of-tree plugin; there is no settings-page card.

Configuration can be overridden in the profile’s cordis.patch.yml via id tool-scout, for example softwareProbeTimeoutMs, softwareCacheTtlMs, maxNames.

- id: tool-scout
  name: '@deepseek-ai/dsh-tool-scout'
  config:
    softwareProbeTimeoutMs: 2000
    softwareCacheTtlMs: 0
    maxNames: 30

Use Cases and Notes¶

Suitable for developers who need DSH agents to read environment facts before reasoning, especially for robot/embedded debugging, multi-platform host checks, and project toolchain checks.

Note: The plugin runs with the permissions of the current dsh process; before installing, you should check the source code, license, and profile configuration.

The verified documentation also lists the following limitations:

Package name, repository path, and README title are inconsistent.
Maintainer information is inconsistent: package.json author is ExDragine, repository path clue is omdsh-dev/dsh-scout.
Installation command only provides the /path/to/dsh-scout local path form, not a registry installation command.
Part of the fields in README and package.json are truncated, unable to confirm the complete scripts.
Before use, refer to the complete files in the repository.

Environment requirements:

node: ^22.19.0 || >=24.0.0
packageManager: pnpm@11.7.0

Dependencies:

peerDependencies:
  @deepseek-ai/cordis ^4.0.1
  多个 @deepseek-ai/dsh-* ^0.1.0-rc.6

dependencies:
  @deepseek-ai/schemastery ^3.18.1

Links¶

GitHub：https://github.com/omdsh-dev/dsh-scout

Directory page clue: https://www.skillhub.cn/plugins/omdsh-dev/dsh-scout（未完全确认）

@deepseek-ai/dsh-tool-scout: DeepSeek Harness Environment Detection Plugin