detecting-correcting-hallucinations-llm-generated

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill parses and outputs corrected code and hallucination reports verbatim, so if the input code contains API keys, tokens, or passwords the LLM will reproduce those secret values in its output (allowing exfiltration).