performance-optimization
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The instructions use standard educational language. No patterns for overriding safety filters, bypassing system constraints, or DAN-style injections were found.
- [DATA_EXFILTRATION]: The skill does not access sensitive local files (e.g., SSH keys, credentials) or perform unauthorized network requests. All referenced engine monitors (e.g., Performance.get_monitor) are standard for performance analysis.
- [REMOTE_CODE_EXECUTION]: The skill contains no instructions for downloading or running external code. The code snippets provided for Godot (GDScript) and Unity (C#) are standard examples of object pooling and memory management.
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes user-provided descriptions of performance issues. However, since the skill does not grant the agent access to autonomous tools or sensitive APIs, the attack surface for processing untrusted data is minimal.
- [OBFUSCATION]: All content is clearly readable. No base64, hex-encoded commands, zero-width characters, or homoglyph attacks were detected.
- [PRIVILEGE_ESCALATION]: The instructions do not involve any commands requiring elevated permissions, such as sudo or modifying system configuration files.
- [DYNAMIC_CONTEXT_INJECTION]: The skill does not use any dynamic shell execution patterns (e.g., exclamation-backtick syntax) to generate content.
Audit Metadata