[中文](README.md) · English
# dsh-douyin-creator
**Make DeepSeek check official Douyin sources before giving creator advice.**
[](https://github.com/forgeturl/dsh-douyin-creator/actions/workflows/ci.yml)
[](package.json)
[](LICENSE)
[](https://github.com/deepseek-ai/deepseek-harness)
### Real model-and-tool end-to-end verification
On 2026-08-30, we completed a real Web UI conversation with **DeepSeek-V4-Flash** in an isolated profile. The model actually invoked `douyin-data-diagnosis`, `douyin_official_search`, and `douyin_official_source_read`, then returned clickable official Douyin sources and the four evidence layers. This was more than a static code or “enabled” status check.
The API key remained only in an isolated local test configuration and was never written to the repository, conversation, or screenshots. See the [acceptance test report](docs/验收测试报告.md) for the full test scope and remaining remote-only boundaries.
## What it helps with
| Your question | What the plugin does |
| --- | --- |
| Is this topic worth making? | Checks recommendation, search, content-quality, and governance sources, then proposes a testable judgment |
| Does this script contain fake algorithm claims? | Reviews platform claims, numbers, causality, cover copy, and spoken copy |
| Why did a post underperform? | Checks content, audience, search, originality, and governance signals instead of blaming an “account score” |
| What should I publish next week? | Turns hypotheses, production tasks, metrics, and stopping rules into a weekly experiment |
| What did Douyin actually publish? | Returns source IDs, titles, dates, official URLs, and relevant excerpts |
| I am starting from scratch. What should I make? | Builds an account card, three content pillars, and three first-week MVP posts |
| I need the original context, not just an excerpt. | Reads the indexed source by ID with pagination and transcript/freshness warnings |
## Real sources, layered conclusions
The image below is not a fabricated product interface. It is an explanation built from real records in the current plugin index, with source IDs and official URLs preserved.
Answers are separated into four layers:
1. **Officially published** — directly supported by an official source.
2. **Reasonable inference** — derived from public mechanisms but not promised by Douyin.
3. **Operational advice** — a testable and reviewable creator action.
4. **Account-specific conclusion** — only valid when the user provides a defined sample, time window, and metric basis.
The plugin must not present fixed traffic-pool ladders, universal account scores, permanent metric weights, universal retention thresholds, guaranteed viral formulas, or a universal best posting time as official Douyin claims.
## Real creator practice
The maintainer uses this workflow to operate the Douyin account 《商道人物志》. As of **August 2026**, the account is approaching **60,000 followers**, with meaningful results in day-to-day practice.
This is one maintainer-operated account, not a promise that other accounts will reproduce the same views, followers, or revenue.
## Two tools and five focused skills
| Name | Purpose |
| --- | --- |
| `douyin_official_search` | Searches 73 knowledge units and 231 chunks, returning source IDs, official URLs, dates, and evidence excerpts |
| `douyin_official_source_read` | Reads source context by ID with pagination and transcript warnings |
| `douyin-creator-onboarding` | Builds a beginner account card, content pillars, and first-week MVP posts |
| `douyin-topic-evaluation` | Evaluates audience needs, search opportunity, content value, and governance risk |
| `douyin-script-review` | Reviews platform claims, facts, numbers, and causality, then provides replacement-ready copy |
| `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 |
You can also invoke a skill explicitly:
```text
/douyin-script-review
```
## Compatibility and data freshness
| Item | Status |
| --- | --- |
| Plugin version | `0.2.0` |
| Node.js | `^22.19.0` or `>=24.0.0` |
| DeepSeek Harness | Install, configuration composition, plugin loading, real model calls, and tool calls verified with `0.1.0-rc.7` |
| Source snapshot | 2026-08-13 to 2026-08-14 |
| Index build | 2026-08-21 |
| Knowledge base | 73 knowledge units; 231 chunks |
| Distributed media | Zero image, audio, or video files in the npm package; README visuals are not downloaded with it |
Questions about current policies, interfaces, or rules must still be checked against the latest official pages rather than answered only from this snapshot.
## Install and develop
Non-technical users should start with the Windows/macOS double-click launchers in the repository root. Developers can use the following command instead (no global `dsh` required):
```bash
npx --yes github:forgeturl/dsh-douyin-creator#main
```
Or use the `npx` invocation recommended by DeepSeek Harness directly:
```bash
npx --yes @deepseek-ai/dsh@0.1.0-rc.7 plugin --profile web add 'github:forgeturl/dsh-douyin-creator#main'
```
Install a local checkout:
```bash
npx --yes @deepseek-ai/dsh@0.1.0-rc.7 plugin --profile web add /absolute/path/to/dsh-douyin-creator
```
The runtime `index.js` is committed, so a Git install does not need a build step. Use the `plugin` subcommands provided by your installed Harness version to update or remove the plugin; run `npx --yes @deepseek-ai/dsh@0.1.0-rc.7 plugin --help` first to confirm the current syntax.
## Download-size boundary
The public npm package contains runtime code, five skills, and text-only knowledge data. The illustrated guide and double-click installers ship with the GitHub/customer ZIP. Repository visuals are limited to bounded SVG/WebP files under `docs/images/`; each must stay below 300 KiB and their combined size below 600 KiB. GIF, video, audio, PDF, and high-resolution PNG/JPG files are rejected. README visuals are excluded from the npm package.
## Verify locally
```bash
npm test
npm run pack:check
```
The checks cover natural-language search, source reading, readable output, index counts, skill metadata, the Harness manifest, one-click installers, media boundaries, and reciprocal README language links.
## Contributing
- Contribution guide: [CONTRIBUTING.md](CONTRIBUTING.md)
- Security policy: [SECURITY.md](SECURITY.md)
- Changelog: [CHANGELOG.md](CHANGELOG.md)
- License: [MIT](LICENSE); see [NOTICE.md](NOTICE.md) for third-party source rights
Official-source additions and focused skill improvements are welcome. Every platform fact must include an official URL, publication date, retrieval date, and source type. Machine transcripts are supplementary evidence only.