---
name: kitnelo-web-upgrade
description: Improve an existing website's task completion, discovery, and SEO with product and user evidence. Use for requested website UX or feature iterations, including bilingual tool directories; keep the site's stack and scope.
---

# KitNelo Web Upgrade

KitNelo authored this instruction-only skill for Codex. It is not an official OpenAI skill and needs no KitNelo account, connector, or API key. Work inside the user's website repository with the tools already available.

## Find the user task

Start from the requested outcome: who is trying to do what, with which input, and what usable result they expect. Inspect the relevant page, route, current implementation, and repository instructions. If access to the live page is available, try the task before editing; otherwise use a local preview and report that scope.

Trace the journey from finding the feature through understanding its prerequisites, entering information, handling an error, and using the result. A page view is not task completion; a button click is not a successful download. Use provided analytics or observed behavior as evidence. Label hypotheses and sample data, and do not invent conversion rates or user research.

Choose the most consequential friction within the user's scope and implement a complete improvement. For an open-ended upgrade, prefer a small usable journey over adding several unfinished tools. Preserve the product's brand, framework, existing content, and user edits unless the request requires changing them.

## Make the journey understandable

- Name actions after the user's outcome. Explain inputs and supported formats before a file or long form is submitted. Show examples when an unfamiliar input prevents people from starting.
- Put a useful primary action near the task. Reveal optional settings when needed; advanced controls should not block the first successful run.
- Errors identify the affected input and a remedy. Preserve valid input and drafts when a retry or route change is possible. If inputs change after processing, mark the old result and prevent exporting it as current.
- Completion gives a usable result and a relevant next step, such as a real download, copy, or handoff. Verify that the advertised capability actually works; do not hide a manual workflow behind an automatic claim.
- For a directory, distinguish a hosted tool from a local skill or developer integration. Explain installation, required accounts, known pricing uncertainty, and the actual launch destination. A listing is not an integration running on the website.
- For multiple languages, keep labels, help, examples, and errors consistent with the route. Honor a deliberate language choice and test first visits separately from returning users.

## Match SEO to actual pages

For changed public routes, inspect the rendered title, description, canonical URL, crawlable internal links, and sitemap entry. Use the production origin configured by the project, rather than a preview domain. Confirm reciprocal language alternates when translated equivalents exist.

Index pages with distinct useful content. Private results, saved searches, and arbitrary filter combinations need a deliberate index policy and consistent robots metadata or response headers. Check real missing routes return 404. If the project prerenders, verify useful content and links without JavaScript. Structured data must describe what users can actually see and use; never invent reviews, prices, or functionality.

## Prove the improvement

Use the repository's relevant checks, then exercise the changed journey with realistic input on a narrow screen and desktop. Check keyboard access and labels, meaningful error recovery, and the final artifact or destination. When files are exported, inspect their contents rather than treating a click as proof. Add a focused regression test when a meaningful calculation, routing rule, or persistence behavior changes.

Report what changed, why it helps the requested task, what was verified, and material limitations. Describe expected user impact as a hypothesis until measured. Publishing, installing services, changing analytics, or sending messages is within scope only when the user has authorized it; an authorized deployment is complete after checking the deployed journey.