Your first 20 minutes with cyberg7-skills

TL;DR

Install the free starter pack — skills land in ~/.claude/skills/.

Don't memorize the catalog; let find-skills surface the right one per task.

Ship one real change with coding-standards + tdd-workflow so the habit sticks.

Why skills, not a longer prompt#

Before the first minute, it helps to know what you're installing. When Anthropic introduced Agent Skills in October 2025, Barry Zhang, Keith Lazuka, and Mahesh Murag defined them as:

"organized folders of instructions, scripts, and resources that agents can discover and load dynamically to perform better at specific tasks."

— Anthropic, Equipping agents for the real world with Agent Skills

The key word is dynamically. Skills use progressive disclosure: the one-line summary costs almost nothing, and the full instructions load only when the task calls for them — like a manual that opens at the table of contents, not all 400 pages at once. That's why a library of 154 skills doesn't bloat your context the way a giant CLAUDE.md would. Anthropic's own best-practices guide makes the same point: for anything "only relevant sometimes, use skills instead. Claude loads them on demand without bloating every conversation."

Minute 0: install the starter pack#

The starter pack is a zip of skills plus a tiny sync helper. Unpack it, run the install step, and the skills land in ~/.claude/skills/ where Claude Code reads them automatically at the start of every session.

You don't wire anything per-project. Open Claude Code in any repo and the skills are already there. Setup happens once, not every morning. (If you're still pasting conventions by hand, read Stop re-teaching Claude your coding style next.)

Minute 5: stop reading the catalog#

There are 154 skills. You are not going to remember them, and you shouldn't try. The catalog is not the interface — the task is.

When you start something — "add auth", "write tests for this module", "set up a deploy" — reach for find-skills. It is the meta-skill whose entire job is matching what you're doing to the skills that already solve it. You describe the work; it points you at the procedure that's already been hardened for that work.

This flips the model. Instead of you learning a library, the library finds you at the moment you need it.

Minute 10: ship one real change#

Pick something small and real. Then use two skills together:

coding-standards makes the change look like your team wrote it — import order, export style, file layout, the test runner you actually use.
tdd-workflow makes it work before it ships — test first, watch it fail, write the minimal code to pass.

The combination is the lesson. One skill governs how the code reads; the other governs whether it behaves. And the model is more than capable of holding up its end: on SWE-bench Verified, real GitHub issues, Claude Opus 4.5 scores 80.9% (the first model over 80%) and Sonnet 4.5 scores 77.2%. Your job in these 20 minutes is to give that capability your context, not to supervise its IQ.

Minute 20: make it travel#

Run the sync step (csm sync) so your skills follow you across machines and stay current when we ship updates. Now every session on every laptop starts from the same baseline — your conventions, your test discipline, already loaded.

That's the first twenty minutes. You didn't memorize a thing; you installed once, let the library find the right procedure, and shipped a change the way you'd want every change shipped.

Where to go next#

Install the free starter pack from the homepage.
Use find-skills on your next task instead of guessing.
When you want more depth, add the full library and explore planning-workflow for bigger features.
Going unattended? Read The Stop-hook loop that ran for six hours before you leave an agent running overnight.

More onboarding material in the Getting started pillar.