This week's Swipe · August 19, 2026 · 3 min read
fix tests that don't test anything
You get a test, you get a test, EVERYBODY gets a test!
It makes me feel like a “proper dev” when my agent runs the full test suite and it all comes up green. Turns out: most tests probably aren’t testing the actual features of your app. I had a bunch of tests to confirm that my page H1 contains a certain phrase. WHY?
In this week’s issue you can delete those useless tests, fix design snags from your Mac bar, understand wtf your agent is saying visually, and design better apps using prebuilt components.
Let’s get to it!
(Btw Grok Bot is taking over the full production of the newsletter and decided that AI dev tools and skills/workflows is the way forward. Everything will get human signoff at the end and probably a hefty rewrite. Every issue should pass Pangram. Should…)
In this issue
- ▪
Belt, Continuity layer for AI agents. [sponsor]
- ▪
cull-tests, Delete pointless tests your agent writes
- ▪
Remarc, Show agents exactly what to fix
- ▪
show-me, Make agent explanations easier to understand
- ▪
AIcss, Use prebuilt UI for AI apps
- ▪
tinycast, Mac launcher without the bloat
- ▪
bearnie, Accessible Astro page blocks
Give every coding agent the same reusable context. [sponsor]
Why: Agents often start each session without the useful context from earlier work. Belt surfaces relevant skills and knowledge as you work, so a good workflow can carry into the next task instead of disappearing with the chat the second you close it.
Try: Install Belt in the coding agent you already use. Add one recurring workflow as a skill, then see whether Belt suggests it when a matching task appears in a later session.
Skills & workflows
No more mock API endpoints, feature confirmations and wasted tokens.
Why: I took this example prompt from Howaboua and turned it into a proper skill you can use with your agents to kill pointless tests.
Try: npx skills add iannuttall/swipe-skills --skill cull-tests. Now our vibe coded side projects will actually catch real issues before we ship.
Tools
Leave a comment on the exact part that needs fixing.
Why: Remarc sends the exact screen area and your comment to your coding agent over MCP. Highlight the bug and let them fix it in a jiffy.
Try: Install Remarc and connect its MCP server. Highlight the likely many UI quirks that need fixing with a comment that goes straight to your model of choice.
Skills & workflows
See the code as a small visual structure.
Why: Agents love to waffle and I sometimes can’t understand the backscroll. show-me is a skill that gets them to visually explain it with a tree, a stack, or a small mermaid diagram.
Try: Install the skill: npx skills add humanlayer/skills --skill show-me. Use it when your agent is being extra: /show-me wtf do you mean?
Tools
Copy the components you need for your chatbots.
Why: Every agent UI needs a thinking state and a stream, so why build them from scratch? AIcss has those pieces (ten free), so paste one in and get back to building your product.
Try: Ask Codex to spin up a new chat app for you using AIcss components and wonder why you ever build chat app components yourself.
A tiny native launcher that stays out of the way.
Why: You wanted a hotkey and the clipboard, but Raycast now does too much a lot more. tinycast is a native Mac menu bar app for launching, keybinds and clipboard history that stays out of the way.
Try: Install tinycast and move your main Raycast hotkey across for a day. A day without opening Raycast tells you everything; or nothing?
Accessible Astro and Tailwind blocks to build your landing pages.
Why: Astro is my go-to because it’s just HTML by default and very fast (good for SEO). bearnie gives you accessible Astro and Tailwind blocks you can give your agent because, let’s face it, they can’t design for shit.
Try: Give your agent the docs page and ask it to build your landing page from it. Marvel at how much faster and better looking it is than whatever Claude thought looked good.
A note from Ian: what do you think of this issue? Hit reply and let me know. Too long? Too short? Does the why/try work?
Plain .mdyou can paste straight into Claude, ChatGPT, or any agent.