Trace Eval Loop

Analyze arbitrary session-exported ATIF (Agent Tool Interaction Format / Agent Trajectory Interchange Format, ATIF-v1.7) trajectories — not Harbor jobs/ and not raw .trace.json — reconstruct intent→tool→observation sequences, and derive evidence-backed improvements for LibrAgent builtin tools, prompts, execution policy, and export/harness contracts. Use for: session ATIF eval loops, exported *_trajectory.json review, tool/harness postmortems from UI export, "ATIF 분석", "AITF 분석", "trace-eval-loop", or "세션 trajectory 개선". Not for Harbor jobs/ BM cycles (harbor-harness-improvement-loop / jobs-trace-analyzer), live Session API E2E (libr-tool-e2e-harness-loop), or raw .trace.json dumps (trace-analyzer).

fritzprix Updated 12 repo stars

File contents

fritzprix/libr-agent commit 4a8f9793bc

Frequently asked questions

npx get-skill add fritzprix/trace-eval-loop