Two skills for Claude Code, free and complete. Both came out of real bugs on shipped products, and the numbers in them were measured rather than estimated.
| Skill | What it does |
|---|---|
randomness-correctness |
Catch shuffles and odds that aren’t as fair as they look |
idea-kill-search |
Kill a bad business idea in an hour instead of three months |
git clone https://github.com/Chibuchibs/free-claude-skills.git
mkdir -p ~/.claude/skills
cp -r free-claude-skills/skills/* ~/.claude/skills/
Claude picks them up next session and decides when to use them. Drop them in
.claude/skills/ instead if you only want them on one project. They’re plain
Markdown, so they also work in the Claude desktop and web apps, or pasted into
any assistant that takes custom instructions.
randomness-correctness existsThis line is in a lot of codebases:
const shuffled = items.sort(() => Math.random() - 0.5);
It is not a shuffle. Sort comparators are expected to be consistent, and this one isn’t, so the engine produces skewed permutations.
Here’s where the correct answer landed across 200,000 runs of a four-option quiz:
| Position | Should be 25% | Actual |
|---|---|---|
| 0 | 25% | 28.3% |
| 1 | 25% | 28.2% |
| 2 | 25% | 24.7% |
| 3 | 25% | 18.8% |
Always guessing the first option beat guessing at random. Anyone who noticed had an edge, and the quiz was quietly unfair to everyone who didn’t.
Check it yourself:
node -e "
const pos=[0,0,0,0], N=200000;
for(let i=0;i<N;i++){
const o=['a','b','c','d'].map((t,i)=>[t,i]).sort(()=>Math.random()-0.5);
pos[o.findIndex(x=>x[1]===0)]++;
}
pos.forEach((c,i)=>console.log(i, (c/N*100).toFixed(1)+'%'));"
The skill has the Fisher–Yates fix, how to test a distribution properly, the trap that makes your test wrong rather than your code, and how to check that any odds you display to users match what the code actually rolls.
One thing worth knowing if you go looking for this bug: it usually appears more than once. In the codebase it came from, it was in eleven places.
idea-kill-search existsThree searches, run before building anything:
<idea> competitors pricing
<idea> free alternative
<idea> regulation <current year>
The third is the one people skip, and it kills ideas the other two never would. A rule change can make a working business unbuildable overnight and leaves no competitors behind to warn you, because they all quit.
A real example from the skill: a payments reconciliation tool looked viable on the first two searches — a few competitors, all enterprise-priced. The third search found that the central bank had introduced a rule removing the data access the product depended on. Dead, in an hour, instead of three months in.
The skill also has four structural checks that catch what searching misses: who owns the pipe and would they ship it themselves, whether the hard part is actually the code, what the real bottleneck is, and what you already have live but unmarketed.
The other fourteen cover deployment, hosting, verification, retention and marketing — things like testing whether your on-screen instructions actually work, pricing a sponsored post by reach rather than member count, and moving hosts without breaking your email.
If these two are useful, those are the same thing. If they’re not, don’t buy it.
MIT. Use them, change them, build on them.