Show HN: Halftrace – See the shape of your agent's failures from your logs

Category: devtools

Tags: agent-diagnostics, llm-logs, compliance-analysis

Score: 7.0/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 6)

Halftrace is a diagnostic tool for agent trajectories that analyzes logs from OpenAI, Anthropic, or LangSmith to classify compliance failure modes (perfect, bimodal, gradient, etc.) without requiring new API calls. It combines novel shape-based analysis with practical suggestions drawn from empirical pilot data, filling a gap for debugging rule adherence in LLM agents.

Target audience: backend devs, ai engineers, ml researchers

Repository: https://www.halftrace.dev/ · Python · MIT

View on Hacker News