Methodology
Skill security grades
Every skill in the index gets a grade from a static scan of its public SKILL.md. 801 of 839 score A; the ones with risky patterns are listed below, with what triggered the score.
What the scan looks at
The scan reads the skill’s public SKILL.md text — the file you would install — and looks for patterns that deserve a second look before you run something an agent will follow:
- Piped shell downloads (
curl … | bash) and other fetch-and-execute patterns. - Recursive deletes (
rm -rf) and destructive file operations. - Dynamic execution (
eval,exec,new Function). - Base64 decoding of embedded payloads.
- Credential access:
.ssh,id_rsa,.env, cloud and API keys. - Safety bypasses:
--dangerously-skip-permissions,--no-sandbox. - Process spawning and outbound network calls.
sudoand world-writable permissions (chmod 777).
Each finding subtracts points from 100. A ≥ 90, B ≥ 75, C ≥ 60, D ≥ 40, F < 40.
What it does not do
It is a static text scan, not a runtime audit. A skill can score A and still behave badly, and a skill can score C for a pattern that is perfectly fine in its context (a browser tool that calls curl on purpose). The grade is a prompt to read the source, not a verdict.
We do not execute skills to grade them, we do not scan bundled script files that are not in the SKILL.md, and we re-scan only when we re-index a skill.
How to read a grade
- A — no risky patterns found in the text.
- B / C — one or more patterns found; open the SKILL.md and check whether they are intentional.
- D — several serious patterns together; read the entire file before installing.
Corrections
If you think a grade is wrong, tell us on the contact page with the skill and the line you are looking at.
Grades across the index
- A801
- B19
- C18
- D1
- F0
Skills below grade A
- D sandbox-agent rivet-dev 44/100 pipes a download straight into a shell; reads credential files or secret env vars; makes outbound network calls
- C sentry-create-alert getsentry 60/100 pipes a download straight into a shell; makes outbound network calls
- C claude-api Anthropic 60/100 reads credential files or secret env vars; executes dynamically built code; makes outbound network calls
- C higgsfield-generate Higgsfield AI 66/100 pipes a download straight into a shell; makes outbound network calls
- C stripe-projects Stripe 66/100 pipes a download straight into a shell; makes outbound network calls
- C canghe-post-to-wechat freestylefly 68/100 pipes a download straight into a shell; makes outbound network calls
- C canghe-post-to-x freestylefly 68/100 pipes a download straight into a shell; makes outbound network calls
- C google-agents-cli-adk-code Google 68/100 pipes a download straight into a shell; makes outbound network calls
- C higgsfield-brandkit Higgsfield AI 68/100 pipes a download straight into a shell; makes outbound network calls
- C higgsfield-marketplace-cards Higgsfield AI 68/100 pipes a download straight into a shell; makes outbound network calls
- C higgsfield-product-photoshoot Higgsfield AI 68/100 pipes a download straight into a shell; makes outbound network calls
- C higgsfield-soul-id Higgsfield AI 68/100 pipes a download straight into a shell; makes outbound network calls
- C higgsfield-websites Higgsfield AI 68/100 pipes a download straight into a shell; makes outbound network calls
- C higgsfield-youtube-thumbnail Higgsfield AI 68/100 pipes a download straight into a shell; makes outbound network calls
- C slack-cli slackapi 68/100 pipes a download straight into a shell; makes outbound network calls
- C tdd-workflow affaan-m 68/100 pipes a download straight into a shell; makes outbound network calls
- C huawei-cloud-find-skills huaweicloud 72/100 asks for sudo; makes outbound network calls; spawns system processes
- C caveman-setup juliusbrussee 74/100 reads credential files or secret env vars; makes outbound network calls
- C security-review affaan-m 74/100 reads credential files or secret env vars; makes outbound network calls
- B neon-ai-gateway Neon 80/100 reads credential files or secret env vars; makes outbound network calls
- B create-slack-app slackapi 82/100 reads credential files or secret env vars
- B canghe-image-gen freestylefly 82/100 reads credential files or secret env vars
- B convex-create-component Convex 82/100 reads credential files or secret env vars
- B durable-objects Cloudflare 82/100 executes dynamically built code; makes outbound network calls
- B golang-continuous-integration samber 82/100 reads credential files or secret env vars
- B neon-object-storage Neon 82/100 reads credential files or secret env vars
- B turnstile-spin Cloudflare 82/100 asks for sudo; makes outbound network calls
- B i-have-adhd ayghri 85/100 recursive force delete
- B capacitor-app-spm-migration capawesome-team 85/100 recursive force delete
- B book-to-skill virgiliojr94 85/100 recursive force delete
- B capacitor-app-creation capawesome-team 85/100 recursive force delete
- B google-agents-cli-workflow Google 86/100 executes dynamically built code; makes outbound network calls
- B security-and-hardening Addy Osmani 88/100 executes dynamically built code
- B google-agents-cli-eval Google 88/100 executes dynamically built code
- B okx-agent-payments-protocol okx 88/100 decodes base64 payloads; makes outbound network calls
- B sandbox-migrate-to-next Cloudflare 88/100 executes dynamically built code
- B sandbox-next Cloudflare 88/100 executes dynamically built code
- B sandbox-stable Cloudflare 88/100 executes dynamically built code
Community notes
Found a broken install? Know a skill we missed?
One line is enough: the skill name, what happened, and the command you ran. We read every note and fix the index.