Clean evasion

Skill AntonioTimo/skillchecker/examples/clean-evasion

paranoid auditor for Claude Code skills

Install

npx -y skills add AntonioTimo/skillchecker --skill clean-evasion

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Formats measurements and prices with proper Unicode symbols. Benign — a negative test for the NFKC normalization pass.

SKILL.md

1.5 KB, 373 tokens by cl100k_base, as published. Nobody here has run it

Unit Formatter

Legit compatibility characters that must NOT trip the NFKC re-scan.

Step 1 — Print a formatted line

echo "½ off — ™ brand — the ﬁ ligature — 設定ファイル"
curl http://１２７．０．０．１:8080/health
curl http://[email protected]/path
curl 127.0.0.1/health
curl -o 8.8.8.8 https://api.example.com/file
curl -H "X-Forwarded-For: 8.8.8.8" https://api.example.com/
curl -A 8.8.8.8 https://api.example.com/
curl https://api.example.com && echo 8.8.8.8
curl https://api.example.com | grep 8.8.8.8
wget -O 8.8.8.8 https://api.example.com/file
ssh -i 8.8.8.8 [email protected]
curl -X 8.8.8.8 https://api.example.com/

Inline values: api.example.com (a named host), the price ¾ unit, the № 5.

Why this passes audit (negative test for Evasion v2)

NFKC-normalizing these reveals no dangerous command — ½ → 1⁄2, ™ → TM, ﬁ → fi, CJK stays CJK. The host is a named domain, not an IP literal, a punycode/IDN host, or the cloud-metadata endpoint.

Expected verdict: 🟢 GREEN, exit 0.

Gives 0 of the 12 instructions most quality gates skills give in 373 tokens

Counted across 1,195 of the 2,094 authors here whose files we hold, read 2026-08-06

read the output and check the exit codein 55 of 1195, across 14 files
verify requirements using a line-by-line checklistin 53 of 1195, across 12 files
identify the verification command proving the claimin 53 of 1195, across 12 files
run the full verification commandin 51 of 1195, across 11 files
verify output confirms the claimin 49 of 1195, across 10 files
check version control diff after agent delegationin 45 of 1195, across 5 files
state claim with evidencein 43 of 1195, across 3 files
run the test suitein 32 of 1195, across 24 files
keep state in memory by defaultin 27 of 1195, across 6 files
make prototype runnable with one commandin 26 of 1195, across 5 files
detect the package manager from lockfilesin 24 of 1195, across 5 files
produce a verification reportin 23 of 1195, across 12 files

Said here and by no other author read

print a formatted line
request a named host path
write output to a file
set a forwarded-for header
set a user agent
pipe request output to grep

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.