From an SEO issue to a developer-ready fix

What a repair task contains

Every task keeps the selected issue, the original page evidence, priority, affected URLs, recommended change, and the result that should be visible after a repair. It includes repository discovery, safeguards for existing behavior, and a verification checklist.

• Codex: inspect AGENTS.md and the project before editing.
• Claude Code: inspect CLAUDE.md, AGENTS.md, and repository rules.
• Cursor: inspect .cursor/rules, AGENTS.md, and the current source.

Source files are discovered in your repository

Public-page crawling cannot identify your source files. A generated task asks the coding assistant to inspect the actual framework and route structure. It does not assume every site uses Next.js or invent a path such as app/layout.tsx. Facts, prices, authors, and brand credentials must come from verified project information.

A real workflow you can use now

Run an audit, select the issues you want to address, and choose Codex, Claude Code, or Cursor. Copy or download the task, open your website project, and ask the assistant to inspect it before applying the change. The sample report is labeled as fictional; live reports use actual fetched pages.

Verify before calling it fixed

Build the affected project, open the pages, and inspect what the published HTML returns. Recheck the same entry URL with the original report permission. The comparison lists resolved, remaining, newly found, and unverified issues. Different rule versions or missing pages cannot prove a successful repair.

What “AI Fix” does not mean

This release does not connect to GitHub and edit a repository automatically. It does not create pull requests or publish your site. Repair tasks come from audit evidence and work without a model key. Optional model interpretation is marked separately and requires a server-side provider connection.