saved-markdown（saved Markdown 搜索）

saved-markdown

Publish immutable, anonymous pages to https://saved.md. Pages can be Markdown (with chart widgets), HTML (sanitized static pages with CSS), or JSX (sandboxed interactive React).

Core principle

Frontmatter decides when to use this skill. Everything below defines how to execute once triggered.

Workflow modes

After triggering, pick the mode that best matches user intent. If unclear, default to one-shot and mention that other modes exist.

Pages are immutable — edits always produce a new URL.

Execution flow (high level)

1. Decide mode
2. Decide starting point (template scaffold or freehand)
3. Decide format (Markdown / HTML / JSX)
4. Generate or transform content
5. Validate
6. Publish (unless local-only)

Starting point decision

1) User explicitly asks for a scaffold

• Read templates/INDEX.md
• Show relevant templates (filter by format if known)
• Let user choose
• Load selected template

2) User wants a scaffold from a screenshot / description

• Follow templates/create-template.md
• Create a new template file under:

- templates/markdowns/ for markdown - templates/htmls/ for html - templates/jsx/ for jsx

• Update templates/INDEX.md
• Then continue like a normal template flow

3) User wants to enhance existing markdown

• Analyze content
• Identify visualization opportunities:

- trends → line - categories → bar - proportions → pie - relationships → scatter

• Add multiple charts where justified
• Keep original structure intact
• Publish as a new page

4) User describes content (no scaffold mentioned)

• Match content type using routing table
• Load corresponding scaffold template
• Generate content from it

5) Nothing matches

• Generate freehand:

- title - structured body - optional footer

Format decision

Critical rule

If using HTML or JSX, always explicitly set contentType. Otherwise the page renders incorrectly.

Content-type routing

Use this when generating structured Markdown or HTML content.

Golden rule

Never invent content to fill sections. Omit anything without real data.

Template scaffold workflow

When a template is selected:

1. Read template file from templates/
2. Show the scaffold skeleton to the user in a code block
3. Ask what to change:

- content - sections - colors

1. Replace placeholders with user data and remove irrelevant sections
2. Validate
3. Publish

If the user already provided full content, skip template browsing and go straight to generation.

Markdown enhancement (charts)

When enhancing markdown:

• Add charts using markdown-ui-widget
• Use multiple charts when data supports it
• Ensure:

- numeric values are plain numbers (no $, %, K, etc.) - units go in labels, not values

Template guide: templates/charts.md

Validation checklist

Before publishing:

• Content under 100 KB
• Correct contentType
• Charts valid (if markdown)
• HTML sanitized (no scripts)
• JSX:

- only allowed imports (react, react-dom) - no external packages

• No placeholder text left

Template guide: templates/validation.md

Publishing

Preferred (script)

python scripts/publish.py --file <path> --content-type <markdown|html|jsx> --title "<title>"

Direct API contract (source-of-truth behavior)

Use POST https://saved.md/api/pages with JSON.

Required/expected payload fields:

• markdown (string): page source body for markdown, html, or jsx pages
• contentType (string): "markdown", "html", or "jsx"

Example payload:

{
  "markdown": "# My Page\
\
Published from skill",
  "contentType": "markdown"
}

For markdown pages, contentType may be omitted, but setting it explicitly is recommended for consistency.

Remixing an existing page (immutable URLs)

When users request edits to an existing saved.md URL:

1. Parse page id from URL
2. GET /api/pages/{id} and read markdown + contentType
3. Apply requested edits to the returned markdown
4. POST /api/pages using the edited markdown and the same contentType
5. Return the new URL (old URL remains unchanged)

Never imply in-place updates are possible.