safety-scan

# Safety Scan Scan content for prompt injection, jailbreak attempts, and unsafe patterns. ## When to use Before processing untrusted input (user submissions, API payloads, webhook data), scan it to detect prompt injection, adversarial content, or policy violations. ## Steps 1. **Quick safety check** — call `mcp__plugin_ruflo-core_ruflo__aidefence_is_safe` with the input text for a boolean safe/unsafe result 2. **Deep analysis** — call `mcp__plugin_ruflo-core_ruflo__aidefence_analyze` for detailed threat classification and confidence scores 3. **Full scan** — call `mcp__plugin_ruflo-core_ruflo__aidefence_scan` for comprehensive multi-layer scanning 4. **Train defenses** — call `mcp__plugin_ruflo-core_ruflo__aidefence_learn` with confirmed threats to improve detection 5. **View stats** — call `mcp__plugin_ruflo-core_ruflo__aidefence_stats` for detection rates and false positive metrics ## Threat categories - Prompt injection (direct and indirect) - Jailbreak attempts - Data exfiltration patterns - Instruction override attacks - Social engineering prompts