When I do the same task three times with an AI agent, I write it down as a “skill”: a file describing the procedure, the sources to check, the expected deliverables and the known pitfalls. After that, a URL is enough to run the task.
The problem with prompts
A good prompt gets lost. You retype it from memory, you forget a step, and the result varies from one run to the next. Optimizing a blog post, for instance, takes about ten checks: Search Console queries, the live page, internal links, quoted figures, title and meta description. Nobody types them all every time.
What a skill is
It’s a Markdown file with a name and a description (so the agent knows when to use it), then the procedure. Mine for optimizing a post contains:
- what to run in parallel as soon as it starts: 90 days of Search Console queries, reading the live page, testing internal URLs;
- the deliverables, always in the same order: SEO findings, table of fixes, points to confirm, internal links, title and meta, full HTML;
- content rules: answer in the first sentence, table of contents, words to avoid;
- figures already checked, with their source, so they aren’t researched every time.
Three skills in use
- Optimize a post: I paste the URL and the HTML, I get the analysis and the corrected version.
- Rebuild a page end to end: diagnosis, copy block by block, visuals, form, writing to WordPress, live check. This one bundles Python scripts.
- Weekly recap of contact requests: I paste the week’s requests, I get Monday’s message, always in the same format.
What it changes
The result no longer depends on how focused I am that day. And the skill improves: whenever I dislike a result, I fix the procedure, not just the answer.
Takeaway
A prompt is a request. A skill is a procedure you maintain. If you’re explaining the same thing to your AI for the third time, write it down.