Tag
A deterministic macOS menu bar app that watches coding agent transcripts (Cursor, Claude Code, etc.), extracts claims using regex patterns, and cross-references them against evidence to flag inconsistencies—all running locally without LLM or API calls.
This paper identifies a silent failure mode in tool-using LLM agents where policy violations occur without tool errors or agent self-reporting. The authors propose and evaluate lightweight deterministic pre-execution gates that significantly reduce such failures in the τ²-bench airline domain.