fix(rich-markdown-editor): strict, provenance-aware markdown paste

Pasting code-like text such as `5 * width * height` was italicized by StarterKit's lenient mark
paste rules. Own emphasis with the strict CommonMark parser instead:

- Disable StarterKit's mark paste rules (`enablePasteRules: false`); markdown paste is handled by
  MarkdownPaste (marked) and rich paste by real HTML tags. Input rules (typing) are unaffected.
- Split the paste gate by provenance: structural markdown always parses (faithful GFM tables and
  escaping), while inline-only marks parse for a plain-text paste but defer to a rich HTML sibling so
  a copied table isn't flattened.

Emphasis (`*_ ** __ ~~ ``) renders; `5 * width`, `*args`, and `snake_case` stay literal — matching
Obsidian/Linear.

Author: waleedlatif1

apps/sim/app/workspace/[workspaceId]/files/components/file-viewer/rich-markdown-editor/markdown-paste.test.ts

@@ -104,21 +104,19 @@ describe('markdown paste', () => {
     ['empty string', ''],
     ['whitespace only', '   \n\n  '],
     ['a bare thematic break (ambiguous — needs another markdown signal)', '---'],
-    ['inline-only italic (single asterisk would false-positive on e.g. *args)', 'an *italic* word'],
-    ['inline-only strikethrough', 'a ~~struck~~ word'],
-    ['inline-only code', 'some `code` here'],
   ])('leaves %s to the default handler', (_label, text) => {
     editor = mount()
     expect(paste(editor, text)).toBe(false)
   })
 
-  // Only structural / unambiguous constructs gate the markdown parse. Inline-only marks that
-  // `looksLikeMarkdown` deliberately omits to avoid false positives — single-asterisk italic
-  // (`*args`), `~~`, single-backtick code — are covered by the Markdown extension's own paste path,
-  // not MarkdownPaste, so they belong to a different test surface.
   it.each([
     ['heading', '# Heading', 'heading'],
     ['bold', 'a **bold** word', 'bold'],
+    ['italic', 'an *italic* word', 'italic'],
+    ['underscore italic', 'an _italic_ word', 'italic'],
+    ['underscore bold', 'a __bold__ word', 'bold'],
+    ['strikethrough', 'a ~~struck~~ word', 'strike'],
+    ['inline code', 'some `code` here', 'code'],
     ['bullet list', '- one\n- two', 'bulletList'],
     ['ordered list', '1. one\n2. two', 'orderedList'],
     ['task list', '- [x] done\n- [ ] todo', 'taskList'],
@@ -132,6 +130,27 @@ describe('markdown paste', () => {
     expect(JSON.stringify(editor.getJSON())).toContain(`"type":"${nodeType}"`)
   })
 
+  it.each([
+    ['italic', 'an *italic* word', '<p>an <em>italic</em> word</p>'],
+    ['strikethrough', 'a ~~struck~~ word', '<p>a <del>struck</del> word</p>'],
+    ['inline code', 'some `code` here', '<p>some <code>code</code> here</p>'],
+  ])('defers inline-only %s to a rich HTML sibling (keeps its structure)', (_label, text, html) => {
+    editor = mount()
+    expect(paste(editor, text, html)).toBe(false)
+  })
+
+  it.each([
+    ['space-flanked asterisks', 'area = 5 * width * height'],
+    ['python args and kwargs', 'def foo(*args, **kwargs): pass'],
+    ['snake_case identifiers', 'call user_name and file_path_here'],
+  ])('does not emphasize %s (strict CommonMark)', (_label, text) => {
+    editor = mount()
+    paste(editor, text)
+    const json = JSON.stringify(editor.getJSON())
+    expect(json).not.toContain('"type":"italic"')
+    expect(json).not.toContain('"type":"bold"')
+  })
+
   it('parses markdown-shaped plain text even when an HTML sibling is present', () => {
     editor = mount()
     const html = '<h1>Title</h1><ul><li>a</li><li>b</li></ul>'

apps/sim/app/workspace/[workspaceId]/files/components/file-viewer/rich-markdown-editor/markdown-paste.ts

@@ -3,11 +3,12 @@ import { Plugin } from '@tiptap/pm/state'
 import { parseMarkdownToDoc } from './markdown-parse'
 
 /**
- * Markdown syntax hints. If pasted plain text matches any of these, it's parsed as markdown rather
- * than inserted literally — so a pasted link, image, badge, list, or heading renders as rich content
- * instead of showing its raw `[text](url)` / `# ` source.
+ * Structural markdown — strong signals the plain text is genuinely markdown (a link, image, badge,
+ * list, heading, blockquote, fenced block, or GFM table). Our parser round-trips these more faithfully
+ * than generic HTML→DOM mapping (GFM alignment, escaping, the `./raw-markdown-snippet.ts` constructs),
+ * so they are parsed even when the clipboard also carries an HTML sibling.
  */
-const MARKDOWN_HINTS: ReadonlyArray<RegExp> = [
+const STRUCTURAL_MARKDOWN_HINTS: ReadonlyArray<RegExp> = [
   /^#{1,6}\s/m,
   /\*\*[^*]+\*\*/,
   /\[[^\]]*]\([^)]+\)/,
@@ -18,21 +19,40 @@ const MARKDOWN_HINTS: ReadonlyArray<RegExp> = [
   /^\|.*\|.*\|/m,
 ]
 
-function looksLikeMarkdown(text: string): boolean {
-  return MARKDOWN_HINTS.some((hint) => hint.test(text))
+/**
+ * Inline marks — weaker markdown signals (`*italic*` / `_italic_`, `~~strike~~`, `` `code` ``) that a
+ * rich HTML sibling encodes just as well. Parsed for a plain-text-only paste (so markdown copied from a
+ * terminal or `.md` source renders), but deferred to an HTML sibling: its presence means the source was
+ * rich, and it may carry structure the plain text can't (a copied table's plain form is tab-separated,
+ * not a `| … |` grid, so parsing it would flatten the table).
+ */
+const INLINE_MARK_HINTS: ReadonlyArray<RegExp> = [
+  /\*[^*\n]+\*/,
+  /_[^_\n]+_/,
+  /~~[^~\n]+~~/,
+  /`[^`\n]+`/,
+]
+
+function hasAny(hints: ReadonlyArray<RegExp>, text: string): boolean {
+  return hints.some((hint) => hint.test(text))
 }
 
 /**
- * Parses pasted plain text that looks like markdown into rich content. Pastes inside a code block
- * are left untouched (code is meant to stay literal).
+ * Parses pasted plain text that looks like markdown into rich content, via the strict CommonMark
+ * parser ({@link parseMarkdownToDoc}, `marked`). Pastes inside a code block are left untouched (code
+ * is meant to stay literal).
+ *
+ * Provenance decides plain-text-vs-HTML: a `text/html` sibling (copied from a browser, Slack, Notion,
+ * GitHub, or this editor) is the signal the source was rich. Structural markdown is still parsed from
+ * the plain-text sibling regardless — our parser is more faithful for GFM tables and escaping. But
+ * inline-only marks are equally expressible in HTML, so when a rich sibling is present we defer to the
+ * DOM path, which preserves structure the plain text can't encode. A plain-text-only clipboard (a
+ * terminal, a code editor, a `.md` file) always parses.
  *
- * A clipboard entry that also carries `text/html` (copied from a browser, Slack, Notion, GitHub,
- * or this editor itself) used to always defer entirely to ProseMirror's generic HTML→DOM mapping,
- * even when the `text/plain` sibling was clean markdown our own parser round-trips more faithfully
- * (GFM table alignment, escaping, the constructs `./raw-markdown-snippet.ts` now preserves). Only
- * defer to DOM mapping when the plain-text sibling *doesn't* look like markdown — an HTML clipboard
- * payload with no markdown-shaped plain-text counterpart (a genuinely rich paste from a word
- * processor, a web page selection, …) still goes through the DOM path unchanged.
+ * The strictness of the parse matters: `marked` follows CommonMark flanking rules, so `*text*` becomes
+ * emphasis but a space-flanked `5 * width * height` stays literal. The editor sets `enablePasteRules:
+ * false` so StarterKit's lenient mark paste rules (which would mangle that expression on either path)
+ * never run — emphasis is owned by this parser on the plain path and by real HTML tags on the DOM path.
  */
 export const MarkdownPaste = Extension.create({
   name: 'markdownPaste',
@@ -46,9 +66,11 @@ export const MarkdownPaste = Extension.create({
             if (!editor.isEditable) return false
             if (editor.isActive('codeBlock')) return false
             const text = event.clipboardData?.getData('text/plain')
-            if (!text || !looksLikeMarkdown(text)) return false
-            // Parse through the chunker (linear) so pasting a large markdown blob can't freeze the
-            // main thread the way the underlying superlinear parse would.
+            if (!text) return false
+            if (!hasAny(STRUCTURAL_MARKDOWN_HINTS, text)) {
+              if (!hasAny(INLINE_MARK_HINTS, text)) return false
+              if (event.clipboardData?.getData('text/html')) return false
+            }
             const doc = parseMarkdownToDoc(text)
             if (!doc.content?.length) return false
             return editor.commands.insertContent(doc)

apps/sim/app/workspace/[workspaceId]/files/components/file-viewer/rich-markdown-editor/rich-markdown-editor.tsx

@@ -250,6 +250,7 @@ export function LoadedRichMarkdownEditor({
   const editor = useEditor({
     extensions: EXTENSIONS,
     editable: isEditable,
+    enablePasteRules: false,
     autofocus: streamingAtMountRef.current ? false : autoFocus ? 'end' : false,
     immediatelyRender: false,
     shouldRerenderOnTransaction: false,

apps/sim/app/workspace/[workspaceId]/files/components/file-viewer/rich-markdown-editor/rich-markdown-field.tsx

@@ -97,6 +97,7 @@ function LoadedRichMarkdownField({
   const editor = useEditor({
     extensions,
     editable: !disabled && !isStreaming,
+    enablePasteRules: false,
     autofocus: autoFocus ? 'end' : false,
     immediatelyRender: false,
     shouldRerenderOnTransaction: false,