Home / Security
Security First

Every skill is scanned before you install it

The agent skills ecosystem has a real security problem. Malicious SKILL.md files can steal API keys, exfiltrate environment variables, and execute arbitrary code — silently. SkillSpot is the only free marketplace that automatically scans every community skill before it goes live.

Our Security Promise

🤖

Automated scanning on every submission

Every community skill PR triggers an automated scan via skill-issue — a static analysis tool with 50+ built-in security rules. No human bottleneck for clean skills.

📋

OWASP Agentic Skills Top 10 review

Complex or first-time-submitter skills get a manual review against the OWASP AST10 checklist — the industry standard for agentic skill security.

📢

Full transparency

Every skill shows its scan status and date. Skills that fail scanning are blocked and logged in our public rejected log. We publish what we find.

What We Scan For

Based on the OWASP Agentic Skills Top 10 framework. Each submitted skill is checked against all of these categories.

💉
Prompt Injection AST-01

Instructions that attempt to override your agent's behavior, system prompt, or stated purpose.

Ignore previous instructionsYou are now a different AIDisregard all prior constraints
🔑
Credential Leaks AST-03

Hardcoded API keys, tokens, or passwords — and instructions to read or transmit credentials from your environment.

AWS access key patterns (AKIA…)Private key blocksInstructions to read .env or ~/.npmrc
👁️
Hidden Content AST-02

Invisible Unicode characters, base64-encoded payloads, or other obfuscation used to hide malicious instructions.

Zero-width space charactersBase64-encoded commandsRight-to-left override characters
📡
Network Exfiltration AST-05

Instructions to send data to external endpoints, webhook listeners, or known data-capture services.

curl/wget with env variablesWebhook.site / RequestBin URLsInstructions to POST user data
⚡
Code Execution AST-06

Unsafe eval(), exec(), or shell injection patterns that could execute arbitrary code on your machine.

eval() / Function() callsBacktick shell executionPiping curl output to sh
🎭
Social Engineering AST-07

Misleading descriptions that don't match actual behavior, or pressure tactics to bypass security warnings.

Disable security featuresSkip validation flagsClaims of special authorization

Common Questions

Does passing the scan guarantee a skill is safe?

No. Automated scanning catches known patterns but can't detect every possible threat. Always review the source code of community skills before installing, and use your own judgment. The scan badge means we checked — not that it's perfect.

What is the SkillJect vulnerability?

SkillJect is a class of attack targeting SKILL.md files used by Claude Code, Codex CLI, and Gemini CLI. Malicious skills embed hidden instructions that trigger unauthorized scripts — stealing API keys, exfiltrating environment variables, or installing malware. All our scans are specifically designed to detect SkillJect patterns.

What does "Publisher Verified" mean?

Publisher Verified skills are published directly by platform partners (Anthropic, Vercel, Cloudflare, Sentry, Expo). These are official skills from the companies that build the tools they integrate with. They bypass automated scanning and carry the full trust of their publisher.

How do I get my skill publisher-verified?

Publisher verification is for official platform partners. If you work at a company that wants to officially distribute skills through SkillSpot, reach out. Community skills submitted by individuals go through the standard automated scan.

Added to wishlist