Foreword¶

DeepSeek Harness (DSH) plugin ecosystem emphasizes “everything is a plugin”: encapsulating retrieval, document reading, task workflow, and other capabilities into plugins, and then integrating them into agent workflows. When creating Douyin self-media content, a common problem is: many operational claims come from experience summaries, but can they be traced back to official public documents?

dsh-douyin-creator is a DSH plugin designed for this type of problem. It organizes Douyin’s official public recommendations, distribution, search, governance, and technical materials into a searchable knowledge base, which is then used for topic evaluation, script review, data diagnosis, and content weekly plans. This plugin is maintained by forgeturl, with the repository address at https://github.com/forgeturl/dsh-douyin-creator. The DSH Community Directory is an independent site with no official affiliation to DeepSeek or Synthflow; it should not be understood as an official app store.

What is it¶

dsh-douyin-creator is a DeepSeek Harness self-media creation and operation plugin based on Douyin official materials.

It primarily solves a few categories of problems:

• Before giving Douyin operational advice, first retrieve traceable official documents.
• Distinguish between official public information, reasonable inference, operational advice, and account conclusions.
• Support beginners in establishing account cards, content pillars, and the first week’s minimum viable content.
• Support topic evaluation, script review, data diagnosis, and content weekly plans.

The verified basic information is as follows:

Core Features¶

Below is an introduction to the main capabilities of this plugin that have been verified.

Official Material Retrieval¶

The plugin provides the douyin_official_search tool, which can retrieve 73 knowledge units and 231 slices. The search results return the material ID, official link, date, and a summary of evidence.

This is suitable for handling similar problems:

• Whether the official documentation publicly explains recommendation, distribution, or search mechanisms.
• Whether a specific governance rule has official material support.
• Whether a specific conclusion is merely an inference based on experience rather than the official original text.

Official Source Reading¶

The plugin provides the douyin_official_source_read tool, which can read the official original context in pages by material ID.

When a summary is insufficient to judge details, you can use it to continue viewing the relevant paragraphs of the original text, and clarify the material ID and official link.

Topic Evaluation¶

The plugin supports evaluating topics against user needs, search opportunities, content value, and governance risks.

It does not simply give a general judgment like “worth doing / not worth doing,” but requires breaking down the judgment into traceable materials, testable hypotheses, and reviewable actions.

Script Review¶

The plugin supports reviewing platform mechanisms, facts, numbers, and causal boundaries, and returns copy that can be directly replaced.

This is suitable for checking for high-risk expressions in scripts, such as writing platform mechanisms as experience claims, writing correlation as causality, or using absolute statements without official material support.

Data Diagnosis¶

The plugin supports using the account’s own baseline to diagnose content, audience, search, originality, and governance signals.

This means it emphasizes obtaining account samples, time windows, and statistical definitions first, and then forming account-level conclusions, rather than directly applying generic metrics.

Content Weekly Plan¶

The plugin supports organizing evidence, hypotheses, publishing actions, and metrics into a weekly experiment plan.

This is suitable for turning a single content publication into a testable experiment: first clarify the hypothesis, then arrange the publishing action, and finally use metrics to determine whether to continue, pause, or adjust.

Getting Started¶

The plugin supports beginners in establishing account cards, content pillars, and the first week’s minimum viable content.

For people just starting an account, it can help break down positioning, content direction, and the first batch of works, rather than directly stacking “viral methods.”

Layered Conclusions¶

The plugin requires answering the distinction between four layers:

1. Official Public: Mechanisms, rules, or product descriptions directly supported by the original text.
2. Reasonable Inference: Derived from public mechanisms but not directly committed to by the official side.
3. Operational Advice: Testable, reviewable topic, script, publishing, or diagnosis actions.
4. Account Conclusion: Can only be formed after obtaining account samples, time windows, and statistical definitions.

This helps reduce the mixing of experience summaries, operational advice, and platform official conclusions.

Installation and Activation¶

Before installing, check the environment. For developer environments, it is recommended to have Node.js ^22.19.0 or >=24.0.0.

Installation for Regular Users¶

Regular users can download and unzip the full project, then double-click the corresponding installation file according to the system:

• Windows: Windows-双击安装并启动.cmd
• macOS: macOS-双击安装并启动.command

The installer will not read or modify API Keys, nor will it change global npm configurations.

Installation for Developers¶

Developers can run:

npx --yes github:forgeturl/dsh-douyin-creator#main

Before executing, you can check the current Harness command syntax:

npx --yes @deepseek-ai/dsh@0.1.0-rc.7 plugin --help

For production or team environments, it is recommended to replace #main with a specific commit SHA to avoid upstream updates causing data drift.

Version Status¶

DeepSeek Harness is still in the development preview phase, and command or plugin interfaces may change in the future. This plugin has been successfully installed, configured, loaded, and verified with real models and tool calls using 0.1.0-rc.7. The API Key in isolated testing only exists in the local isolated test configuration and has not been written to the repository, conversations, or screenshots.

Typical Usage¶

Below are verified examples that can be used directly.

First Question¶

You can paste the following into the conversation:

Please first use douyin_official_search to search for "recommendation algorithm content quality",
then distinguish between official public information, reasonable inference, and operational advice to evaluate my topic.

This prompt requires the plugin to first retrieve official materials and then answer in layers of official public information, reasonable inference, and operational advice.

Explicit Skill Invocation¶

You can also explicitly invoke the Skill:

/douyin-script-review

The verified Skills include:

• douyin-creator-onboarding: Helps beginners establish account cards, content pillars, and the first week’s minimum viable content.
• douyin-topic-evaluation: Evaluates topics against user needs, search opportunities, content value, and governance risks.
• douyin-script-review: Reviews platform mechanisms, facts, numbers, causal boundaries, and returns copy that can be directly replaced.
• douyin-data-diagnosis: Uses the account’s own baseline to diagnose content, audience, search, originality, and governance signals.
• douyin-weekly-plan: Organizes evidence, hypotheses, publishing actions, and metrics into a weekly experiment plan.

Applicable Scenarios and Notes¶

This plugin is suitable for creators and operators who want to reduce “experience-post style conclusions” and hope to base Douyin content decisions on material evidence. It is particularly suitable for the following types of tasks:

• Judging before topic selection whether official materials support a claim.
• Reviewing high-risk expressions in scripts before publishing.
• Checking item by item when reviewing data based on content, audience, search, originality, and governance signals.
• Creating a weekly experiment plan rather than scheduling based on feeling.

Please note before use:

• The plugin runs with the permissions of the current DSH process. You should check the source code, dependencies, license, and installation commands before installing.
• The license for this plugin is MIT; the rights notice for third-party public materials is in NOTICE.md.
• The material snapshot is 2026-08-13 to 2026-08-14, and the index was built on 2026-08-21. When questions involving “current, latest, today, current entry, current policies” arise, you must continue to check the latest official web pages; you cannot rely solely on the local snapshot.
• DeepSeek Harness is still in the development preview phase, and command or plugin interfaces may change in the future.
• The maintenance results of a single account do not represent all accounts, and do not constitute guarantees for plays, follower growth, or revenue.
• The currently verified materials do not provide directory page URLs, so this article does not list specific directory page addresses; the DSH Community Directory is an independent site with no official affiliation to DeepSeek or Synthflow.

Conclusion¶

The value of dsh-douyin-creator does not lie in directly providing “viral answers,” but in breaking down Douyin creation and operational judgments into retrievable official materials, explainable inference boundaries, and executable operational experiments. It is suitable for topic evaluation, script review, data diagnosis, and weekly planning, and is also suitable for separating official public information, reasonable inference, operational advice, and account conclusions before giving a final verdict.

GitHub Repository: https://github.com/forgeturl/dsh-douyin-creator