ENG-1832 Add Obsidian eslint to the project

[trangdoan982]

eslint-plugin-obsidianmd catches the plugin-guideline violations that get plugins rejected from the Obsidian community store, but wiring it in naively drowns the signal: its recommended set reports 168 new findings in apps/obsidian, of which only a handful can actually cost us the store listing. This adds a dedicated gate that reports only store blockers, and fixes the 15 real violations so the gate can be blocking from day one.

Replaces the closed #1188, which was blocked on ESLint 8. That's moot now — the repo is on ESLint 9.39.5 and typescript-eslint 8.67.0, so the plugin's peers are satisfied with no catalog scoping.

Review focus

• packages/eslint-config/obsidian-store.js — the gate. Keeps only obsidianmd's own error-level rules and drops everything its recommended set drags in (eslint-comments, depend, import, security, sdl), which contributed 43 findings + 8 unused-directive errors unrelated to store review.
• apps/obsidian/src/components/canvas/DiscourseToolPanel.tsx — the riskiest change. The drag ghost's 11 inline style writes became two class constants. Note the className on the JSX must stay a constant expression; the reactor writes it imperatively and React only clobbers props whose rendered value changed.
• .github/workflows/ci.yaml — new job is blocking. Intentional, but it's a new way for PRs to go red.

Decisions worth challenging

• The gate is not built on @repo/eslint-config. That config loads eslint-plugin-only-warn, which patches ESLint globally on import and forces every rule to "warn" — the exact distinction this gate exists to make. I verified an explicit "error" severity does not override it. Hence a separate config in a separate process. The alternative — dropping only-warn repo-wide — would turn 112 existing Obsidian warnings and 538 Roam warnings into hard errors, so it was rejected as out of scope.
• It lives in packages/eslint-config, not apps/obsidian. It must register the preferArrows / react-hooks / turbo namespaces, or existing eslint-disable comments in our source raise Definition for rule ... was not found. That package already depends on all three; apps/obsidian does not.
• scripts/ is exempt. All 22 no-console errors were in build and test-data tooling that never ships in the plugin bundle, so Obsidian never reviews it. Exempting it took the gate from 37 errors to 15, all in shipped code.
• Only error-level rules are wired in. The 62 warn-level findings (prefer-window-timers, prefer-create-el, ui/sentence-case, …) are style advice, not store blockers. Deliberately deferred — 46 are auto-fixable and worth a follow-up ticket.
• Tailwind over styles.css for the fixes, per project UI guidelines. No new CSS was needed.
• pnpm lint is untouched. base.js and react-internal.js are unmodified, so Roam and the website are unaffected.

Verification

• lint:store exits 0. Negative test: adding el.style.color = "red" is caught as an error with exit 1 — confirming severity survives, which is the whole point of the separate config.
• check-types exits 0; build reports 0 errors.
• Main apps/obsidian lint unchanged at 112 messages / 0 errors, before and after.
• Roam lint 538 messages / 0 errors, no obsidianmd/* rules leaked.
• prettier --check clean on all changed files.
• Runtime, in the live Obsidian app over CDP: injected the new class set and the inline styles it replaced side by side and compared getComputedStyle. All 8 properties identical — display: flex, position: fixed, pointer-events: none, left: 0px, top: 0px, width: 50px, height: 50px, align-items: center — and .hidden resolves to display: none. (First run showed 3 mismatches purely from stale plugin CSS; reloading the plugin resolved all 3.)

Not covered: no unit tests — the change is lint config plus CSS-equivalent refactors, and the repo has no test setup for either. The drag-ghost interaction was verified by computed-style equivalence, not by performing an end-to-end drag on a canvas, so a manual drag is worth a reviewer's minute.

Known limits

• The gate's rule list is derived from obsidianmd's recommended config at runtime, so a plugin upgrade that promotes a rule to error will start failing CI without a config change here. That's the intended behavior, but it makes eslint-plugin-obsidianmd upgrades worth reading closely.
• eslint-plugin-obsidianmd wants @eslint/js@^9.30.1 while packages/eslint-config pins ^9.29.0. Install succeeds and nothing misbehaves, so I left the pin alone rather than widen scope.

🤖 Generated with Claude Code

---

[linear-code]

ENG-1832

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
discourse-graph	Ready	Preview	Aug 22, 2026 6:41pm

[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 ↗︎.

[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

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 1 additional finding.