Foreword¶

In the DSH plugin context, Chinese writing and polishing need to revolve around the genre, the original text, and the final draft. dsh-humanizer is a writing assistance plugin, not an AI detector; it does not score, detect, profile, or allocate quotas. Its approach is to let the model fully read the theory before starting to write, then enter creation or polishing mode, and write in one go.

What This Is¶

dsh-humanizer is maintained by DEEP-IOS and is MIT licensed. It provides Chinese writing and polishing workflows, allowing the model to fully read all the theory before starting to write, become the author through thinking, and then write in one go.

Core Features¶

• humanize_study(genre, mode): Returns the full text of all twenty-one chapters of theory and three sample essays in different styles before writing begins.
• humanize_guard(original, final): Only verifies the retention of content anchors and textual integrity, with no scoring, detection, or profiling.
• humanize_reference(chapter): Consult a specific chapter individually.
• humanize_profile(text): A retired compatibility stand-in that only returns content anchors.
• Supports “write this chapter with humanizer” to enter creation mode.
• Supports “polish this text with humanizer” to enter polishing mode.
• Writing assistance, not an AI detector.

Installation and Activation¶

First, install the plugin into the web profile:

dsh plugin --profile web add "github:DEEP-IOS/dsh-humanizer"

Restart the web after installation. The bundle layer stack is composed at boot time, and changes to the Node half require a restart to take effect.

Runtime requirements are as follows:

engines: node >=18
packageManager: pnpm@10.30.3

Typical Usage¶

1. Creation mode: Send “write this chapter with humanizer.”

2. Polishing mode: Send “polish this text with humanizer.”

3. Use humanize_study(genre, mode) before writing to obtain the full text of all twenty-one chapters of theory and three sample essays in different styles.

4. To consult a specific chapter individually, use humanize_reference(chapter).

5. To verify the original text against the final draft, use humanize_guard(original, final), which only checks the retention of content anchors and textual integrity.

Applicable Scenarios and Notes¶

Suitable for Chinese writing and polishing scenarios, not for use as an AI detector. It does not output probabilities, nor does it identify authors; it does not require submitting text to any external detection; it does not promise any detection score.

Citations, data, formulas, and sources must not be fabricated by the model.

This plugin runs with the permissions of the current dsh process. Check the source code and license before installation.

Links¶

GitHub: https://github.com/DEEP-IOS/dsh-humanizer

Directory page: https://www.skillhub.cn/plugins/DEEP-IOS/dsh-humanizer