-
Failed Assertion Explanations (#2)
- Detailed reasoning for why assertions failed
- Suggestions for fixing the issues
-
Result Caching (#3)
- Speed up local execution by caching assertion results
- Avoid redundant LLM calls for unchanged code
-
Aggregation Modes (strict/balanced/relaxed) (#4)
- Strict: requires unanimous agreement across all evaluations
- Balanced: uses majority voting
- Relaxed: requires only one positive evaluation
-
File Reference in Assertions (#6)
- Allow referencing files in assertions
- Handle potentially large files efficiently
-
CPU-Optimized Model (#5)
- Small fine-tuned model focused on local analysis
- Optimized for CPU execution
- Faster and cheaper inference