Verification system that catches coding agents falsely claiming task completion. Runs 4 parallel checks (file integrity, test quality, scope narrowing, optional LLM judge) over task+claim+diff and returns a weighted 0-100 confidence score with evidence.
verification asyncio agents github-actions pydantic fastapi ai-evaluation openrouter coding-agents code-review-automation llm-judge test-quality confidence-scoring scope-detection agent-overclaim
-
Updated
May 21, 2026 - Python