The Agent Skills Directory

[PROMPT_INJECTION] (SAFE): The skill includes specific templates (agent-guardrails-input.ts) that demonstrate how to implement guardrail agents to detect and block prompt injection and jailbreak attempts.
[DATA_EXPOSURE] (SAFE): No hardcoded credentials were found. Templates (e.g., api-realtime-route.ts) explicitly warn against exposing the OPENAI_API_KEY to the client-side and provide patterns for secure server-side proxying and ephemeral key generation.
[EXTERNAL_DOWNLOADS] (SAFE): The scripts/check-versions.sh script uses npm view to check for SDK updates from the official npm registry. This is a standard, non-malicious development utility.
[COMMAND_EXECUTION] (SAFE): The skill implements tools using safe patterns. High-stakes actions like account deletion or payments are implemented with a 'Human-in-the-loop' (HITL) requirement in agent-human-approval.ts, requiring explicit manual confirmation before execution.

openai-agents