Make the Obsidian plugin loadable on mobile

[trangdoan982]

Reviewer brief

• Result: the Obsidian plugin bundle no longer contains any Node or Electron code, and ships with isDesktopOnly: false. This is the load-blocker MVP only — it makes the plugin start on mobile so a designer can evaluate it via BRAT. Touch affordances are explicitly not in scope.

• Review focus: two behavior swaps that also affect desktop users.

  • splitFrontmatter + parseYaml replaces gray-matter. Semantics are matched by unit test, with one deliberate difference: gray-matter threw on malformed YAML, this returns empty frontmatter instead. That is closer to how Obsidian itself tolerates bad YAML, but it does mean a bad import file now yields a node with no frontmatter rather than a hard failure.
  • The markdown-link regex now captures a leading ! instead of using a lookbehind. Worth confirming the image-vs-link split still reads correctly at importNodes.ts.

• Risk or follow-up: verified statically and by unit test, not on a physical device. The touch gaps below are known and unaddressed, so the plugin is usable-but-incomplete on mobile:

  • The "Create <NodeType>" button lives inside a hover-only tooltip (tagNodeHandler.ts) — unreachable on touch.
  • The canvas context menu (TldrawViewComponent.tsx) is driven by tldraw's right_click; long-press does not produce it, so "Convert to discourse node" is unreachable on canvas.
  • Wikilink drag-to-canvas (wikilinkDragHandler.ts) uses HTML5 drag events, which do not fire on touch.
  • The single-character node-tag hotkey has no non-keyboard trigger.

  The 17 addCommand registrations all surface in the mobile command palette, which is the reliable entry point in the meantime.

The most durable change here is dropping Node builtins and electron from the esbuild external array. Marking them external is what let three require() calls survive into the shipped bundle in the first place; without it esbuild fails the build instead. This is recorded in apps/obsidian/AGENTS.md so it is not quietly reverted.

Verification

• Production bundle contains require() for only obsidian and @codemirror/view — previously also fs (gray-matter), path (mime-types), and buffer (js-yaml). 2.2MB.
• Zero regex lookbehind assertions in the bundle. The one remaining (?< is a core-js feature probe using a named capture group.
• Remaining process.env / Buffer references are all guarded — tldraw wraps its env reads in try/catch, and both Buffer uses are behind typeof Buffer !== "undefined".
• pnpm install --frozen-lockfile then pnpm ci:validate from the repo root: pass, including the 15 new Obsidian unit tests.

apps/obsidian had no test runner; this adds vitest mirroring the Roam app's setup and exposes it through test:unit so root validation picks it up.

Loom video

Scope check

• Ran $scope-check against the ENG ticket and final diff.
• Scope beyond Done When: no ENG ticket — this branch came from a direct request to unblock a mobile BRAT build for design evaluation.

Local delegated full review

• Ran a comprehensive review of the entire final diff in a subagent with a fresh context.

🤖 Generated with Claude Code

[supabase]

This pull request has been ignored for the connected project zytfjzqyijgagqxrzbmz because there are no changes detected in packages/database/supabase directory. You can change this behaviour in Project Integrations Settings ↗︎.

---

Preview Branches by Supabase.
Learn more about Supabase Branching ↗︎.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
discourse-graph	Skipped		Sep 3, 2026 4:38pm UTC

[graphite-app]

PR size/scope check

This PR is over our review-size guideline.

• Recommended: ~200 lines changed
• Acceptable limit: up to 400 lines when well-scoped/self-contained
• Preferred file count: fewer than 5 files

Please split this into smaller PRs unless there is a clear reason the changes need to land together.

If keeping it as one PR, please add a brief justification covering:

• What single problem this PR solves
• Why the files/changes are coupled

[graphite-app]

When YAML parsing fails, the function returns body: content (the original full content), but this is inconsistent with the successful parse path. When splitFrontmatter succeeds but parseYaml throws, the frontmatter block markers (---) are already stripped from body, so returning the original content re-introduces them.

Impact: A file with malformed YAML frontmatter will have the raw frontmatter block (including --- delimiters) left in the body text, which will render as markdown content instead of being hidden.

Fix: Return the stripped body instead:

catch {
  return { frontmatter: {}, body };
}

This way, the malformed frontmatter block is removed from the body (matching the successful parse behavior), and users get empty frontmatter instead of a parsing error (the desired tolerance behavior mentioned in the PR description).

Suggested change

	} catch {
	return { frontmatter: {}, body: content };
	} catch {
	return { frontmatter: {}, body };

Spotted by Graphite



Is this helpful? React 👍 or 👎 to let us know.