openai-security-best-practices
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's instructions in SKILL.md under the '# Overrides' section create a surface for indirect prompt injection by explicitly directing the agent to prioritize instructions found within the analyzed project's documentation and code that may contradict the security best practices defined in the references directory.
- Ingestion points: As part of its workflow, the agent reads the project codebase and 'specific rules and instructions in the project's documentation and prompt files' which are untrusted inputs.
- Boundary markers: The skill lacks instructions to treat embedded project instructions as untrusted, instead advising the agent to 'not fight with them' when an override is encountered.
- Capability inventory: The agent has access to powerful tools including Bash, Read, Grep, Glob, Write, and Edit, which allow it to modify the filesystem and generate reports based on potentially adversarial instructions found in the code it reviews.
- Sanitization: The instructions do not define any validation or sanitization process for the untrusted override instructions, potentially allowing an attacker to use code comments to suppress security findings.
Audit Metadata