The Agent Skills Directory

[SAFE]: The skill serves as a professional educational resource and template library for the OpenAI Agents SDK (JavaScript/TypeScript). It provides structured guidance on building multi-agent workflows, voice interfaces, and framework integrations.
[INDIRECT_PROMPT_INJECTION]: The skill templates define an attack surface for indirect prompt injection as they are designed to ingest untrusted user input for processing by LLMs. However, the skill explicitly addresses this risk by providing detailed implementation patterns for both input and output guardrails, safety filters for PII and harmful content, and human-in-the-loop (HITL) approval mechanisms to validate high-stakes tool executions.
[COMMAND_EXECUTION]: The provided shell script (scripts/check-versions.sh) performs standard version checks on dependencies using npm view. These operations are limited to identifying package metadata for compatibility purposes and do not involve executing untrusted code or exfiltrating sensitive data.
[CREDENTIALS_UNSAFE]: The skill demonstrates best practices for secret management. It instructs users to use environment variables for API keys and provides a dedicated template (templates/nextjs/api-realtime-route.ts) showing how to generate secure, ephemeral session tokens for client-side applications to avoid exposing primary API keys to the browser.

openai-agents