conradcaffier03

Skill Audit: can Claude actually read your skills?

Claude Code puts a short list of your skills into every conversation: each skill's name plus its description. Claude uses the description to decide which skill fits your request.

That list has a budget: 1% of the model's context window (Claude Code docs, "Skill descriptions are cut short"). When your skills don't fit, Claude Code keeps every name but drops descriptions, starting with the skills you use least. A skill without a description has no words Claude can match against, so it only runs when you type /its-name.

My setup on 30 Sep 2026: 118 skills, 59,112 characters of descriptions, a 30,000 character budget.

conradcaffier03

🔷 SHAPE — the structure pass your humanizer skips

Freebie for the SHAPE keyword (reel-80, "Your humanizer is fixing the wrong thing"). Deliver as a public GitHub Gist. By the end you have a Claude skill that checks the shape of a draft (how it is built), not the words, and proposes at most three structural changes.

---

What I tested

conradcaffier03

The Silent Thread Check

Every inbox has two kinds of finished conversations. The ones that ended because they were done, and the ones that ended because you stopped writing. They look identical. Nothing in your inbox marks the difference — which is why the second kind quietly piles up.

This finds the second kind. It takes about ten minutes, and you do not need to install anything or know what an API is.

---

conradcaffier03

📊 SKIPRATE — the two-number check for your own reels

Freebie for the SKIPRATE keyword (reel-79C, "71 reels. Seven worked."). Deliver as a public GitHub Gist. By the end you can look at any reel you have posted and say, in under a minute, whether it earned distribution or whether Instagram was right to stop showing it — using two numbers that are already in your app.

---

The finding this comes from

conradcaffier03

✅ VERIFY — the check that stops a scraper from handing you made-up numbers

Freebie for the VERIFY keyword (reel-79B, "It read a page that doesn't exist"). Deliver as a public GitHub Gist. By the end you have a 30-line script and a Claude skill that refuse to pass you a number that isn't actually on the page.

---

What happened to me

conradcaffier03

🎬 RENDER — turn an HTML file into a real MP4

Freebie for the RENDER keyword (reel-79A, "This video was a text file"). Deliver as a public GitHub Gist. By the end you have a folder on your machine where an HTML file renders to a finished vertical MP4 — and an AI agent that can edit that file for you.

---

Why this exists

conradcaffier03

🆓 Find out what to post about — the last-30-days setup

You don't need more content ideas. You need to know what already worked in your niche this month — what people actually upvoted, watched and argued about. This setup makes Claude read the last 30 days of Reddit, YouTube, Hacker News and GitHub for you and rank it by real engagement. Free, open source (MIT), no API key needed to start. The tool is last30days by mvanhorn — I just tested it and wrote down what it really does. Given away free by @buildwith.conrad.

What I measured (09 Sep 2026, my own machine, no key, no account): one question — "AI video tools" —

conradcaffier03

Control any app with Claude — the free setup

Free. Open source. About 5 minutes.

Claude can't click buttons in your apps. But it can talk to them through a command line — and there's a free tool that writes that command line for you.

It's called CLI Anything (Apache-2.0, 48,000+ stars on GitHub). Point it at an app, and Claude reads the app's code, writes the commands, and installs them.

conradcaffier03

Web Research Setup — let Claude read a hundred websites for you

Free. No credit card. No terminal. 2 minutes.

Claude can read one website. It can't read a hundred — until you connect this. Firecrawl is the open-source tool (174,000+ GitHub stars) that lets Claude search the web, open real pages and hand you back clean answers instead of 14 open tabs.

---

conradcaffier03

🆓 The Four Rules — one file that fixes how your agent codes

Your agent guesses instead of asking. It writes 400 lines where 40 would do. It "improves" code you never mentioned. These four rules are one file you drop into your project, and the behaviour changes on the next prompt. No install, no dependency, no tool. Given away free by @buildwith.conrad.

---

Where this comes from