Windsurf Prompt Checklist: What I Check Before Sending a Task

Why I use a checklist before Windsurf

When I first started using AI coding tools, I often sent Windsurf a rough idea and waited for the output. That worked for simple screens, but it created problems on real projects. The tool might create the right page in the wrong folder, update generated output instead of the source generator, or add a CTA that did not match the current routing system.

A checklist slows me down for five minutes. That sounds like friction, but it usually saves more time later. Before Windsurf touches the project, I want to know the page goal, the target files, the constraints, the internal links, the validation commands, and the expected final report.

This is especially useful on a static SEO site. A page can look fine in the browser while the sitemap, canonical URL, hreflang tags, or language version is wrong. The checklist keeps the task tied to the real publishing pipeline.

The checklist I use

My Windsurf checklist starts with the user goal. I write what the reader or user should be able to do after the change. For a blog article, the goal might be to learn a workflow and move to the main guide. For a bug fix, the goal might be to stop a code block from overflowing on mobile.

Next I list the source of truth. If the page is generated, I say clearly that the source generator should be changed, not only the generated HTML. If the route must appear in GitHub Pages, I mention both `site_output` and `docs` so the tool does not stop too early.

Then I add safety rules. My common rules are: do not deploy, do not call external APIs, do not create fake affiliate links, do not add secrets, do not enable auto-posting, and do not break `/go/` redirect behavior.

Finally, I include the test commands. The prompt should not end with just "make the change." It should end with build, sync, validation, language integrity, go-page checks, and a short report. That makes the result easier to trust.

Prompt example: create a supporting article

When I ask Windsurf to create a supporting article, I do not only give the title. I describe the cluster, the main article, the internal links, and the content quality rules. This prevents the article from becoming an isolated page.

The prompt below is the kind of instruction I would send when building another page in this AI coding workflow cluster.

Create a supporting SEO blog article for /blog/chatgpt-prompts-for-windsurf/. Keep the current site style and bilingual pipeline. Add clear H1, intro, table of contents, practical sections, prompt examples, common mistakes, FAQ, related links, and CTA to the workflow checklist. Link naturally to /blog/chatgpt-prompts-for-windsurf/, /windsurf-review/, /comparisons/cursor-vs-windsurf/, and /free-ai-coding-workflow-checklist/. Do not add fake affiliate links, API calls, or auto-posting logic. Rebuild, sync docs, and run validation.

Prompt example: fix a specific issue

For bug fixes, I try to make the prompt narrower. Windsurf performs better when the task has a visible problem, a small set of likely files, and a definition of done.

This kind of prompt is useful when a page looks broken, but I do not yet know which CSS rule or generator function is responsible.

The blog page /blog/chatgpt-prompts-for-windsurf/ has long prompt examples inside pre/code blocks that overflow outside the article card. Inspect the shared article CSS in modules/site_builder.py and any generated blog page output. Fix the source CSS so pre, code, .prompt, and .code-block stay inside the content width on desktop and mobile. Do not change article content. Rebuild, sync docs, validate the site, and report the exact files changed.

Common prompt mistakes

The biggest mistake is asking Windsurf to do strategy, writing, coding, SEO, tracking, and deployment cleanup in one vague instruction. It can try, but the result is harder to review.

The second mistake is not naming the files or pipeline. If a project generates `docs` from `site_output`, the prompt should say that. If Vietnamese pages are generated from English pages, the prompt should mention the localization pipeline.

The third mistake is skipping validation. A page is not done because it exists. It is done when links work, sitemap is updated, language integrity passes, and the page reads naturally.

Related guides in this workflow

If you want the full process, start with my guide to ChatGPT prompts for Windsurf. If your problem is bilingual output, read the mixed-language guide next. If Windsurf creates the first version but the project still needs careful cleanup, use the Windsurf-to-Codex workflow.

These articles are meant to work together. One explains the prompt habit, one gives a checklist, one handles language problems, and one explains when to move from first build to focused repair.

FAQ

Next step

If you want to use this process on your own static site or app idea, start with the main workflow article, then use the checklist before sending your next prompt to Windsurf.