The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates for building AI agents that ingest untrusted user input and external tool outputs, which constitutes a surface for indirect prompt injection.
Ingestion points: Request body processing in templates/nextjs/api-agent-route.ts, templates/cloudflare-workers/worker-text-agent.ts, and templates/cloudflare-workers/worker-agent-hono.ts.
Boundary markers: The skill includes a dedicated template for implementing an Input Validator agent in templates/text-agents/agent-guardrails-input.ts to detect and block injection attempts.
Capability inventory: The agents are designed to perform network operations via the OpenAI API and execute user-defined tools, such as the weather and search tools shown in templates/text-agents/agent-basic.ts.
Sanitization: The skill demonstrates the use of InputGuardrail and OutputGuardrail components to validate, filter, and sanitize content, which is a recommended mitigation strategy.
[EXTERNAL_DOWNLOADS]: The skill references legitimate, well-known libraries and SDKs.
Dependencies: SKILL.md and templates/shared/package.json recommend installing @openai/agents, @openai/agents-realtime, and zod via npm. These are official packages from a trusted organization.

openai-agents