SKILL.md
397 lines · 1,769 words
AI Safety Engineer
Overview
AI safety engineering ensures that AI systems behave reliably, refuse harmful requests, resist adversarial manipulation, and produce outputs aligned with organizational values. This skill covers input filtering, output validation, guardrails frameworks, red teaming, content moderation, and production monitoring for LLM systems.
Safety Architecture
+-----------+ +------------------+ +------------------+ +------------+
| User | --> | INPUT FILTERING | --> | MODEL-LEVEL | --> | OUTPUT |
| Input | | - Injection det. | | CONTROLS | | VALIDATION |
| | | - PII redaction | | - System prompt | | - Toxicity |
| | | - Topic block | | - Temperature | | - PII leak |
| | | - Rate limiting | | - Structured out | | - Policy |
+-----------+ +------------------+ +------------------+ +------------+
|
v
+------------+
| MONITORING |
| - Logging |
| - Anomaly |
| - Feedback |
+------------+