openai-sdk-security-scan
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by design, as it is intended to ingest and process user-provided source code for security auditing. This code is considered untrusted and could theoretically contain instructions designed to influence the agent's output.
- Ingestion points: User-supplied code files provided during a "review", "audit", or "scan" request.
- Boundary markers: The skill provides a structured rules table (OAI-KEY-001 to OAI-RETRY-001) to constrain the agent's analysis criteria.
- Capability inventory: The skill body consists of markdown instructions and code examples for detection; it does not invoke external scripts or perform file-write/network operations.
- Sanitization: The skill instructs the agent to report findings based on a specific scoring schema.
- [EXTERNAL_DOWNLOADS]: The skill includes references to external resources for educational and documentation purposes.
- Fetches configuration and SDK guidelines from OpenAI's official GitHub repositories.
- Links to OpenAI's developer platform for documentation on Structured Outputs and Function Calling.
- References security guidelines from the official OWASP LLM Top 10 project website.
Audit Metadata