Introduction

The DeepSeek Harness (DSH) plugin ecosystem currently consists mainly of English content. For the compliance requirements of the Mainland China market (such as Advertising Law absolute terms, official document formatting standards, and PIPL privacy protection), the community lacks a complete set of automation tools. This causes teams to repeatedly encounter pitfalls or leads to incidents in release processes due to missed compliance or engineering details.

feely0208/dsh-cn-skills is an engineering and compliance skill set for Chinese-language scenarios. It contains 7 independent skills, covering external content compliance, official document formatting, pre-release self-checks, AI content labeling, personal information protection, app store pre-checks, and key scanning. All skills are offline static checks; they do not access the network, do not upload data, and only read local files.

Core Features

The plugin set includes the following 7 core skills:

  1. External Content Compliance Red Line Review
    Checks internet-facing wording, Advertising Law absolute terms, absolute promises, and the risk of exposing internal files to the public network.
  2. Chinese Official Document Formatting
    Generates documents that comply with the formatting requirements of the GB/T 9704-2012 national standard.
  3. Pre-release Self-check
    Checks deliverable manifest reconciliation, version consistency, architecture correctness, and other items to ensure release actions are complete and accurate.
  4. AI-Generated Content Labeling Compliance Self-check
    Cross-references the “Measures for Labeling AI-Generated Synthetic Content” and national standard GB 45438—2025 to verify explicit/implicit labels.
  5. Personal Information Protection Cross-check
    Cross-references the Personal Information Protection Law (PIPL) to check permission usage, third-party SDK lists, and user rights entry points.
  6. App Store Listing Pre-check
    Checks app filing, privacy policy accessibility, permission usage scenario descriptions, and minor protection measures.
  7. Key and Sensitive Information Scanning
    Scans for accidentally committed cloud provider AKs, tokens, private keys, and other items to prevent sensitive information leakage.

Installation and Enablement

The installation command is:

dsh plugin --profile web add @deepwhale-cn/dsh-bundle-cn-skills

dsh-bundle-cn-skills is a combined package that declares dependencies and mounts the above 7 skills at once. After installation, restart DSH, and the Agent will automatically load these skills in relevant conversations.

If you need to install a skill individually, you can use:

dsh plugin add @deepwhale-cn/dsh-cn-compliance

Typical Usage

This skill set applies to the following specific scenarios:

  • When producing external content output: when modifying the official website, writing public documentation, publishing WeChat Official Account articles or in-app copy, perform compliance red line review.
  • When drafting formal documents: generate official documents, notices, or reports that comply with the GB/T 9704-2012 standard.
  • During the release process: when releasing, uploading installation packages, switching download links, or creating a Release, perform pre-release self-checks.
  • When developing AI products: when building AI products, designing export features, or preparing for app store listing, verify AI labeling compliance.
  • When preparing for privacy and compliance: when preparing for listing, modifying the privacy policy, adding permissions or SDKs, perform PIPL cross-checks.
  • When committing or handing over code: before committing, open-sourcing, releasing, or handing over, scan for keys and sensitive information.

Notes

  • Offline operation: all skills are fully offline static checks; they do not access the network, do not upload data, do not write to your project, and only read local files.
  • No built-in sensitive word list: the plugin does not include a politically sensitive word list, avoiding becoming a carrier of sensitive content. The compliance review portion requires manual review.
  • False positive avoidance: one design principle of the tools is to reduce the false positive rate (avoid misclassifying technical terms or code comments as violations). Some hits in scan results (such as technical terms in code) are by design and need to be judged based on context.