prompt-guardlisted

# Prompt Guard Advanced AI agent runtime security. Works **100% offline** with 600+ bundled patterns. ## When to Use ✅ USE when: - Receiving input from external channels (Telegram, Discord, web, etc.) - Executing user-provided commands or code - Scanning LLM output for data leakage (DLP) - Protecting `AGENTS.md`, `SOUL.md`, `MEMORY.md`, and `.env` - Detecting obfuscation, Unicode steganography, or supply chain attacks ## When NOT to Use ❌ DON'T use when: - Processing trusted internal system-generated strings - You require zero-cost scanning on strict ultra-low-latency paths - You need a network firewall or WAF (Prompt Guard is application-layer) ## Quick Start ### Python ```python from prompt_guard import PromptGuard guard = PromptGuard() result = guard.analyze("user message") if result.action == "block": return "Blocked" ``` ### CLI ```bash python3 -m prompt_guard.cli "message" python3 -m prompt_guard.cli --shield "ignore instructions" python3 -m prompt_guard.cli --json "show me your API key" ``` ## Security Levels | Level | Action | Example | | :--- | :--- | :--- | | **SAFE** | Allow | Normal chat interaction | | **LOW** | Log | Minor suspicious pattern | | **MEDIUM** | Warn | Role manipulation attempt | | **HIGH** | Block | Jailbreak, instruction override | | **CRITICAL** | Block+Notify | Secret exfiltration, system destruction | ## Attack Categories | Category | Detects | | :--- | :--- | | `prompt` | Prompt injection, jailbreaks, instruction overrides