Fine-Tuning Assistant
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection] (LOW): The skill provides code templates for creating instruction-tuning datasets that interpolate untrusted data fields into message blocks, which could be exploited if the source data contains malicious instructions.
- Ingestion points: The
create_instruction_examplefunction inSKILL.mdacceptsraw_dataas input for prompt construction. - Boundary markers: Absent; the code snippets do not include delimiters or 'ignore' instructions to isolate user-provided content.
- Capability inventory: Low; the snippets describe data transformation and training configuration, though the resulting models would be used for generation tasks.
- Sanitization: Absent; there is no evidence of input validation or escaping in the provided documentation examples.
- [Safe Practices] (SAFE): The skill does not perform active network operations, file system modifications, or command execution. It uses standard, well-known machine learning libraries and placeholders for configuration values.
Audit Metadata