modelscope-publish-learn - SKILL.md Agent Skill

name: modelscope-publish-learn description: | Publish articles to ModelScope.cn 研习社 (Learn/Study section). Use this skill whenever the user wants to write, create, edit, or publish an article on ModelScope's learning community. Handles the full workflow from article creation to metadata editing to final publication via browser automation and API calls.

ModelScope 研习社文章发布

Prerequisites

Ensure kimi-webbridge daemon is running (~/.kimi-webbridge/bin/kimi-webbridge status). If not, start it.
User must be logged in to ModelScope in their browser.

Workflow Overview

1. Create draft  →  POST /api/v1/articles (or click "下一步" on /learn/create)
2. Edit content   →  PUT /api/v1/articles with ContentDraft
3. Edit metadata  →  Navigate to /learn/editor/{id}
4. Publish        →  Click "发 布" then confirm

Key API Endpoints

Endpoint	Method	Purpose
`/api/v1/articles`	POST	Create new article draft
`/api/v1/articles`	PUT	Update article (title, desc, content, metadata)
`/api/v1/articles/{id}`	GET	Read article data
`/api/v1/articles/{id}/publish`	PUT	Publish article

Critical Field Mapping

PUT payload uses Description for the article summary/blurb
GET response returns it as Desc
Always include ALL fields in PUT — partial payloads clear missing fields

ContentDraft Format

The editor uses a custom JSON array format. Each paragraph:

["p",{},["span",{"data-type":"text"},["span",{"data-type":"leaf"},"PARAGRAPH TEXT"]]]

Full document wrapper:

["root",{},["p",{},[...]],["p",{},[...]],...]

Encoding Safety (Windows)

NEVER pass Chinese characters through inline bash/python -c commands on Windows — the console uses GBK and will corrupt UTF-8.

Always use the helper script with file input or write arguments to a file first:

python ~/.kimi/skills/modelscope-publish-learn/scripts/gen_payload.py \
  --id 433474 \
  --title "文章标题" \
  --desc "文章简介" \
  --content-draft 'JSON_STRING' \
  --subjects '["数据集"]' \
  --domains '["人工智能"]' \
  --image-url "COVER_IMAGE_URL" \
  --output ~/.kimi/skills/modelscope-publish-learn/scripts/payload.json

Then send via browser fetch with TextDecoder:

const b64 = 'BASE64_FROM_FILE';
const bytes = Uint8Array.from(atob(b64), c => c.charCodeAt(0));
const json = new TextDecoder().decode(bytes);
fetch('https://modelscope.cn/api/v1/articles', {
  method: 'PUT',
  headers: {'Content-Type':'application/json'},
  credentials: 'include',
  body: json
});

Metadata Page Form Fields

Navigate to https://modelscope.cn/learn/editor/{id}

Field	Selector	How to Fill
Title	`input#Title`	Use `fill` tool
Description	`textarea#Description`	Use native setter + dispatch input event (see script)
Subject	`input#Subjects`	Type and press Enter
Domain	`input#Domains`	Type and press Enter
Cover	Click default image	`img[src*="learn_blue.png"]` etc.

Use scripts/fill_metadata.py to programmatically fill all metadata fields.

Publish Button Sequence

Click "发布" (usually index 6 in the button list)
Wait for confirmation modal
Click "知道了" (usually the last button)
Page redirects to /my/myspace?activeTab=learn

Common Pitfalls

The rich text editor (modelscope_ding_editor_) is canvas-like and resists all standard automation — use the API for content instead.
The PUT /api/v1/articles endpoint does full replacement, not partial update.
The listing page (/my/myspace) may cache old data briefly; the article detail page (/learn/{id}) is the source of truth.
On Windows, Python printing Chinese characters to the console may look garbled due to GBK encoding, but the actual bytes in the generated files are correct UTF-8. Trust the file output, not the console display.