v0.3.0 — OmniRank now knows which file is wrong 🎯

[bemoshiur]

🎯 v0.3.0 — OmniRank now knows which file is wrong

---

Every other SEO tool keys its output to a URL, because a URL is the only thing it can see. OmniRank runs inside your repository, so it resolves

https://yoursite.com/pricing
  → app/(marketing)/pricing/page.tsx
  → the missing alternates.canonical
  → a three-line diff

That's the whole idea. Screaming Frog, Semrush and Ahrefs can't follow — not because they're slow, but because they've never seen your source.

⚡ Try it

omnirank audit https://your-site.com          # what's wrong
omnirank fix   https://your-site.com          # which file, and the diff

fix writes nothing in 0.3.0. It prints the diff and stops. Applying arrives in v0.4.0, and that split is deliberate — the locator is the riskiest component in the project, so it ships and gets proven before anything earns write access.

🧠 The idea underneath

auto_fixable — a boolean set by four findings, all in one module — is gone. It reflected which file a finding lived in, not whether its fix was safe. Two orthogonal axes replace it:

🏷️	fixTier — what the correct edit needs: mechanical · templated · drafted · advisory · infrastructure
🛡️	applicability — whether this instance may be applied: safe · unsafe · display-only

applicability is a min() over the tier ceiling, how confidently the file was located, the edit's blast radius, and a hard cap on protected surfaces. Every input can demote; none can promote.

Because safety isn't a property of the rule. Adding a canonical is trivial on a single-route file and catastrophic on a layout serving 10,000 routes. Same rule, opposite answer, decided by context only a tool inside your repo can see.

Only 4 of 48 findings have a generator. That's the intended answer, not a timid one — OpenRewrite's first principle is "we favour making fewer changes over making wrong changes."

🐛 What review caught

Worth saying plainly, because it's the part that took the longest. Seven real bugs were found before release, each reproduced against real input:

• The locator resolved /blog to a _private folder with exact confidence — a URL Next.js 404s on.
• The same bug for @slot folders, which also made it fail to locate the homepage of Next.js's own official parallel-routes example.
• The static resolvers escaped the project root: ../OUTSIDE resolved above the repo.
• A JSON-LD fix produced a patch that applied cleanly and deleted the node it claimed to repair.
• Every diff against a CRLF file failed to apply.
• The canonical was inserted before <!doctype html> on minified pages, forcing quirks mode.

None were found by reading diffs. All came from feeding the thing awkward, real-world input.

💬 Tell us

• Run omnirank fix on your repo — did the located file match reality? This is the one thing we most need real-world data on.
• Which framework? Next.js App Router, static, Jekyll and Hugo are supported today. Astro, Nuxt, SvelteKit return unknown on purpose rather than guessing — say which you'd want next.
• Did it decline where you expected a fix? A display-only verdict you disagree with is a useful bug report.

_{Full notes in the CHANGELOG · Docs · Wiki}