These are the frameworks behind my client work. Not prompt tricks. Actual staged methods, with the reasoning that makes them work, free to take and run.
Anyone can run a tool. The hard part is knowing what to ask it.
Most AI resources are a prompt with a personality. They work once. Then the situation shifts and they fall apart, because there was never a method underneath.
What I publish here is the method. The stages, in order, and why that order. The questions each stage forces you to answer. The failure mode each one is designed to catch.
The files are real. I run these on paid work. You get the same version I do.
Six methods, documented end to end.
Most content plans start with search volume. That's the wrong end of the problem. This one starts with the questions a buyer actually asks across five stages, then checks how AI answers them today and where you're absent.
Learn the methodologyDesign feedback usually stalls at taste. Four passes fix that. Hidden Truth, What If, Feel, and Combine. You come out with one idea the project commits to and an emotional target you can test against.
Learn the methodologyYour buyers already wrote your copy. It's sitting in forum threads and review complaints. This pulls their exact language and turns it into a messaging framework instead of a word cloud.
Learn the methodologyAn AI visibility audit tells you where you're invisible. A keyword tool tells you what people search. Run separately, they both underperform. Joined, they produce a four-week calendar with a reason behind every slot.
Learn the methodologyA crawl returns 400 issues. Nine of them move anything. After 100+ audits, this is the pass I use to find which nine and ignore the rest.
Learn the methodology"Make it modern" produces slop. This forces the opposite. Named aesthetic vocabulary, decomposed references, hard spec values, and a critique pass held separate from the build.
Learn the methodologyThese are not prompts. Each one is a formalized framework shipped as a SKILL.md specification — staged methodology, failure modes, and output contracts — versioned in the public repo and built for production use. Import one into Claude and run it.
Grant Writing
The full library lives at github.com/Phinneas/salish-sea-skills. More specs ship as the remaining skills are documented.
What these are not.
They're not turnkey. Each one assumes you can judge the output and push back on it. They're built for my stack and my clients, so some steps will need reshaping for yours.
They also won't tell you what your business should be. That part is still yours.
Why I give these away.
I don't sell files. I sell judgment applied to a specific situation, which is the part that doesn't travel in a download.
Publishing the method is also the fastest way to show you how I think. Read one. If the reasoning holds up, you already know what working together looks like.
