Foreword

The plugin ecosystem of DSH emphasizes “everything is a plugin,” and the community directory is also an independent site, not an official app store from DeepSeek or幻方. In scenarios where code review is conducted within a dsh session, temporary prompts often lead to issues such as unstable review scope, inconsistent severity levels, and inconsistent output formats.

The 7DGroup team provides 7dgroup-ai/dsh-skill-7d-code-reviewer. It standardizes the code review process, severity levels, scoring dimensions, knowledge base loading methods, and report templates for AI-assisted code review in any dsh session.

What This Is

7dgroup-ai/dsh-skill-7d-code-reviewer is a template-driven code review skill plugin designed for DeepSeek Harness (DSH).

Verified basic information is as follows:

  • Name: 7dgroup-ai/dsh-skill-7d-code-reviewer
  • Maintainer: 7dgroup-ai
  • Category: Workflow
  • License: MIT
  • Version: 0.1.0-rc.5
  • Prerequisites: dsh CLI, Node ^22.19.0 || >=24.0.0, pnpm 10+
  • Peer dependencies:
  • @deepseek-ai/cordis ^4.0.1
  • @deepseek-ai/dsh-invariants ^0.1.0-rc.5
  • @deepseek-ai/dsh-skill ^0.1.0-rc.5

It is installed via a composite bundle without modifying the DSH core. Uninstallation is done by removing the corresponding bundle line, and it does not come with executable scripts attached to the skill.

Core Capabilities

Template-Driven Review

The plugin separates “what to review” and “how to present it”:

  • SKILL.md determines the review content and severity levels
  • templates/ is used solely for presentation

The HTML report template uses a pure placeholder approach, where all placeholders must be filled, and dynamic values need HTML escaping.

Five-Step Review Process

The review is executed in five steps:

  1. Receive the task
  2. Quick scan
  3. Line-by-line review, loading references/ as needed
  4. Severity level classification
  5. Generate the report

Three-Level Severity Levels

Severity levels are categorized as:

  • critical
  • medium
  • minor

Each level includes repair suggestions.

Four-Dimensional Scoring

Scoring covers four dimensions:

  • Code quality
  • Security
  • Performance
  • Maintainability

Each dimension is scored from 1–10, and an overall assessment and summary are generated.

Dual Output

The plugin outputs both:

  • Text summary
  • HTML report

The HTML report is saved as:

code-review-report-{timestamp}.html

Built-in Knowledge Base

The built-in knowledge base is loaded on demand, including:

  • Coding standards
  • Security checklists: SQL injection, XSS, authentication/authorization, sensitive data leakage
  • Review examples

This allows reference materials to be loaded when needed, rather than being stuffed into the prompt.

Installation and Enabling

Prerequisites before installation:

dsh CLI
Node ^22.19.0 || >=24.0.0
pnpm 10+

Installing in a dsh Session

Below is an example of requesting installation within a dsh session:

安装插件 github:7dgroup-ai/dsh-skill-7d-code-reviewer

Equivalent English request:

Install the plugin github:7dgroup-ai/dsh-skill-7d-code-reviewer

Installing with dsh CLI

The following command uses GitHub shorthand and is suitable for direct installation:

dsh plugin --profile <name> add github:7dgroup-ai/dsh-skill-7d-code-reviewer

Equivalent full URL form:

dsh plugin --profile <name> add git+https://github.com/7dgroup-ai/dsh-skill-7d-code-reviewer.git

Here, --profile <name> corresponds to the profile started with dsh --profile <name>.

Git Installation and pnpm allowBuilds

Git installation triggers its own prepare build. pnpm will block build scripts for git dependencies until they are allowed in the profile’s pnpm-workspace.yaml.

If the first add fails, first write the exact package key printed by pnpm into:

~/.dsh/profiles/<name>/pnpm-workspace.yaml

Then retry the installation. Example configuration:

allowBuilds:
  '@7dgroup/dsh-skill-7d-code-reviewer@git+https://github.com/7dgroup-ai/dsh-skill-7d-code-reviewer.git#<sha>': true

If the github: shorthand is used during installation, the package key will be different. Always use the exact package key actually printed by pnpm.

It is recommended to pin a commit, for example:

...#<sha>

This avoids changes in later pushes altering the code actually run during installation.

Regarding the npm name @7dgroup/dsh-skill-7d-code-reviewer: It is only valid after the package is published, and verified information has not confirmed its publication status. It is recommended to use the GitHub spec for installation first.

Typical Usage

After installation, you can initiate review requests regarding code quality issues within a dsh session. According to verified information, applicable scenarios include:

  • Pre-commit code review
  • Pre-MR code review
  • Security audits of existing code
  • Quality assessment before refactoring
  • Enforcement of team coding standards
  • Code quality issues in dsh conversations

The plugin provides text summaries following the template-driven process and generates HTML report files.

Applicable Scenarios and Notes

This is suitable for users who:

  • Need to obtain stable, formatted code review results within a dsh session
  • Require fixed inspection dimensions before commits or MRs
  • Need to reuse coding standards, security checklists, and review examples
  • Want to transform the review process from free-form prompts into a templated workflow

Notes before use:

  • The plugin runs with the permissions of the current dsh process
  • When installed via Git and with builds allowed, it runs the package code outside the agent sandbox
  • Check the source code, license, and installation scripts before installation
  • Use pinned commits to reduce risks from changes in later pushes
  • License is MIT, version is 0.1.0-rc.5
  • Whether the npm package has been published is not confirmed in verified information; prioritize installation via GitHub spec

Directory and Repository

Plugin directory page:

https://www.skillhub.cn/plugins/7dgroup-ai/dsh-skill-7d-code-reviewer

GitHub repository:

https://github.com/7dgroup-ai/dsh-skill-7d-code-reviewer

The core value of this plugin is templating the code review process: the review scope, severity levels, scoring dimensions, and report output all have a fixed structure, making it suitable for integration into dsh workflows.