Intent-Code Divergence
- Category
- Not specified by scanner
- Confidence
- 96% confidence
- Finding
The skill contains contradictory scope statements: it repeatedly instructs the agent to curate, rank, and order items by stated criteria, but later says the agent does not 'rank, or judge a list.' This ambiguity can cause downstream agents or orchestrators to ignore intended safety boundaries, either letting the model make unverified evaluative claims or causing humans to overtrust rankings as if they were already vetted. In a content-generation workflow focused on 'best of' lists, that inconsistency increases the chance of misleading rankings, undisclosed bias, or fabricated comparative judgments slipping into published output.
- Content
