The finished version
Checkpoint 3 is where this skill ends up after more iteration. Read it before you load it. It adds three things on top of what you built:
- Audience detection. The same findings, written differently depending on who will read them.
- A scoring guide. What a 9 means versus a 4, so scores mean the same thing run to run.
- Example findings. One finding written two ways, so the model sees the shape you want instead of guessing it.
Repo Roast track
Repo Roast adapts to a developer (blunt, file names and line counts), an engineering manager (business cost and prioritization), or a new teammate (a welcoming lay of the land). Ask for a roast “for my manager” or “for onboarding” and watch it switch.
Read checkpoint 3, or load it:
./setup.sh repo --checkpoint 3Draft Roast track
Draft Roast adapts to who reads the draft: customers (hardest on buzzwords and vague claims), coworkers (hardest on buried ledes and missing asks), or decision makers (hardest on missing numbers, owners, and dates). Say “this goes to leadership” and watch it switch.
Read checkpoint 3, or load it:
./setup.sh draft --checkpoint 3setup.sh saves your version as SKILL.md.bak first. Copy it back if you prefer yours.
Compare with a neighbor
Pair up. Run your skills on the same input and put the outputs side by side.
- What constraints did they add that you didn’t?
- Whose findings would you actually act on?
- What’s one line you’d steal?
Install it for good
Repo Roast track
setup.sh already linked your skill into ~/.claude/skills/repo-roast (and ~/.agents/skills/repo-roast for Codex), so it works in every repo. The link points at the kit folder. To keep the skill after you delete the kit, copy it instead:
./setup.sh --cleanup
cp -R skills/repo-roast ~/.claude/skills/repo-roastUse ~/.agents/skills/ for Codex.
Draft Roast track
Terminal. setup.sh already linked your skill into ~/.claude/skills/draft-roast (and ~/.agents/skills/draft-roast for Codex). The link points at the kit folder. To keep the skill after you delete the kit, copy it instead:
./setup.sh --cleanup
cp -R skills/draft-roast ~/.claude/skills/draft-roastThe scripts read draft.md in whatever folder you start your agent from.
Web or desktop. Save your skill as SKILL.md inside a folder named draft-roast, zip the folder, and upload the zip in Claude’s skills settings. Or start from the starter zip and paste your version over its SKILL.md. Once it’s uploaded, you can ask “roast this draft” in any chat without pasting the skill.
What comes next
You built one skill well. The pattern works for any task you repeat:
- Gather evidence (scripts, or pasted context).
- Close off what must never happen (constraints).
- Define the shape of the answer (structure).
- Work in phases.
- Define what “good” means (self-assessment).
skill-creator can scaffold the next one. Ask Claude to help you create a skill and describe the task. Now that you know what a good skill looks like, you can judge what it hands back.
Starter pack
A few skills I use, so you can see what good ones feel like. All of these need Claude Code (a terminal).
| Skill | Use it to | Call it with |
|---|---|---|
| ideation | Turn a messy idea into a plan you can build from | /ideation:ideation |
| grill | Get grilled with questions until your plan holds up | /grilling |
| wizard | Generate a step-by-step script for setup only a human can do | /wizard |
| brainrot | Get through a long doc as narrated TikTok-style sludge | /brainrot |
| explain-this | Actually understand a paper, article, or codebase, with quizzes | /explain-this |
| socratic-tutor | Get quizzed until you understand a change you just made | /essentials:socratic-tutor |
| tighten-prose | Clean AI tells out of your writing | ask “tighten this up”, or /content:tighten-prose |
ideation is a plugin. In Claude Code:
/plugin marketplace add nicknisi/ideation
/plugin install ideation@ideationsocratic-tutor and tighten-prose come from my plugin marketplace:
/plugin marketplace add nicknisi/claude-plugins
/plugin install essentials@nicknisi
/plugin install content@nicknisigrill, wizard, brainrot, and explain-this live in github.com/nicknisi/skills. Copy the folders you want into your skills directory:
git clone https://github.com/nicknisi/skills.git ~/nicknisi-skills
cp -R ~/nicknisi-skills/grill ~/nicknisi-skills/wizard ~/.claude/skills/Skills from a plugin get the plugin’s name as a prefix, like /ideation:ideation.
Open each SKILL.md before you use it. You know how to read one now. Notice which ones set disable-model-invocation: true, and ask yourself why.
Your turn
Name one task you’ll explain to an AI more than once this week. Write it down with:
- What you keep re-explaining.
- What evidence it needs.
- One thing it must never do.
That’s the first draft of your next skill.