<aside> 🔧 The builds behind the stories — the scripts, configs and patterns from the newsletter, written up so you can copy them. Secrets stripped; this is the reusable version. New ones land as I make them.
</aside>
The story: Reading the API instead of the bill found me £180 a year
What it does. Instead of clicking through a billing dashboard, I asked my hosting provider's API for the whole inventory in one call — every domain, certificate and add-on. Six paid SSL certificates at ~£30/yr each, plus a DNS add-on I'd forgotten I was paying for: about £180 a year, staggered across the calendar so no single month's bill ever looked wrong. Then I curl'd each site and confirmed every one was already serving a free Cloudflare or Let's Encrypt certificate — the paid ones weren't even in the chain.
The reusable move (you don't need a hosting API). Export CSVs — from your bank, your SaaS providers, wherever money quietly leaves — and have your LLM scan them weekly. Annual charges are the ones that get you: they land once, in a month you're not watching, and stay invisible the rest of the year. A whole-year view catches what twelve monthly glances never will. Then verify each line against reality before you cancel anything.
Copy this prompt:
Here's a CSV of my [bank / SaaS / hosting] charges for the last 12 months.
1. List every recurring charge, monthly and annual, grouped by what it's for.
2. Flag anything that renews annually — those are easy to miss.
3. Flag anything that looks redundant or that I might no longer use.
Don't tell me to cancel anything yet — just show me what's there so I can check each one.
What it is. A small bundle of Claude Code skills, dropped in all at once instead of hunting them down one repo at a time. They follow one honest sequence — the difference between chatting with AI and actually building with it:
Think first → write without the slop → in your own voice → turn mistakes into rules.
think-first — stops you and applies a mental model before a big decision: architecture, pricing, what to build next.antislop — detects and fixes the tells that make writing sound AI-generated, with an editor mode that fixes them in place.voice-analyzer — reads your writing samples and builds a portable voice profile (a VOICE.md). The setup step for voice-editor.voice-editor — edits AI output to match that voice profile, so drafts sound like you, not the model.lessons-learned — runs a structured retrospective after something breaks, and encodes the fix so it doesn't happen twice.Get it. Public on GitHub, MIT-licensed, yours to take apart:
git clone <https://github.com/aplaceforallmystuff/signal-starter-kit.git>
cd signal-starter-kit
./install.sh
Requires Claude Code. Restart your session, then try /think-first, /antislop, /voice-analyzer, /voice-editor, /lessons-learned.