mayguard
por balkanblbn
Coding Agents & IDEsResumo
A security auditor for agent skills.
Descrição
O que esta skill faz
Static analysis tool that scans agent skill directories for security threats before installation. Detects credential theft patterns, suspicious network calls, destructive commands, and code obfuscation. Returns a safety rating (SAFE, CAUTION, SUSPICIOUS, or DANGEROUS) with specific findings per file.
Catches attack patterns specific to agent skill architectures, like env file exfiltration and eval-based obfuscation, that general-purpose linters miss.
Quando usá-la
- Auditing a downloaded skill before moving it to the active skills folder
- Detecting hardcoded API keys or env file access in third-party skills
- Identifying outbound webhook or tunnel calls in untrusted skill code
- Scanning community-shared skills before running them locally
- Checking for privilege escalation or destructive commands in a new skill
Exemplo de fluxo de trabalho
User asks: Audit the skill I downloaded to ~/Downloads/new-skill before I install it
- 1
Run
python3 scripts/audit.py ~/Downloads/new-skillagainst the target directory - 2
Scan source files for credential access patterns including .env and SSH key references
- 3
Check for suspicious network calls such as webhook endpoints and ngrok or localtunnel usage
- 4
Detect obfuscation techniques including eval, exec, and base64 decoding
- 5
Compute a numerical risk score and assign an overall safety rating
Report showing status CAUTION, risk score, and a list of flagged patterns with the specific files and lines that triggered each warning
Instalação
clawhub install balkanblbn/mayguardExecute o comando da CLI ou cole o prompt no Claude Code, Cursor, Codex ou no agente de código que você usa.