agent-guardrail

Add guardrails to an agent: $ARGUMENTS

Before Starting

1. Search OpenAI docs for guardrail patterns:
   • Use mcp__openai-docs__search_openai_docs with query "Agents SDK guardrails" for SDK-native guardrail support
   • Use mcp__openai-docs__search_openai_docs with query "Agents SDK input validation" for request validation patterns
2. Read the existing guardrail implementation:
   • src/agents/shared/guardrails.ts — nonEmpty, maxLength, contentFilter, noErrors implementations
   • src/agents/shared/enhanced-agent-bridge.ts — applyInputGuardrails() and applyOutputGuardrails() in the bridge pipeline
3. Read the database schema for guardrail persistence:
   • src/lib/database/schema.ts — search for agent_guardrails and agent_guardrail_assignments tables

Guardrail Types

Input Guardrails

Validate user messages BEFORE they reach the agent. Return { passed: boolean, reason?: string }.