All skills, agents, and integrations in Magic Powers are automatically scanned for security issues identified by the Snyk ToxicSkills study (2026), which found 13.4% of community skills contain critical vulnerabilities.
| # | Check | Description |
|---|---|---|
| 1 | Hardcoded secrets | API keys, tokens, passwords embedded in files |
| 2 | Dangerous shell commands | rm -rf /, sudo, chmod 777, pipe-to-bash |
| 3 | Prompt injection | "Ignore previous instructions" and similar patterns |
| 4 | Env variable exfiltration | Dumping process.env, os.environ, printenv to external targets |
| 5 | Write to sensitive paths | Modifying ~/.bashrc, /etc/crontab, /etc/sudoers, etc. |
| 6 | Crypto wallet/private key | Wallet files, seed phrases, private key hex strings |
| # | Check | Description |
|---|---|---|
| 7 | Broad file access | Access to ~/.ssh, ~/.aws, ~/.kube, /etc/passwd |
| 8 | Data exfiltration | POST requests to external URLs |
| 9 | Eval/exec calls | Dynamic code execution patterns |
| 10 | Encoded payloads | Large base64 blobs that could hide malicious content |
| 11 | Malicious package install | npm install / pip install from URL/git (not registry) |
| 12 | Network recon | nmap, netcat, port scanning tools |
| 13 | Obfuscated strings | Hex/unicode escape sequences hiding payloads |
| 14 | Overly permissive instructions | "No restrictions", "bypass safety", jailbreak enablers |
| Directory | Files | What |
|---|---|---|
skills/ |
SKILL.md |
All workflow skills |
agents/ |
*.md |
All agent definitions |
integrations/ |
*.md, *.mdc |
Cursor, Copilot, Aider, Windsurf, etc. |
hooks/ |
* |
Session hooks and scripts |
commands/ |
*.md |
Slash commands |
bash scripts/security-audit.shSecurity audit runs automatically on:
- Every push to
mainthat modifies content files - Every pull request
- Weekly scheduled scan (Monday 00:00 UTC)
If you find a security issue, please email security@pmai.space instead of opening a public issue.