reasoning-trace-optimizer

Analyzing reasoning traces from AI agents detects failure patterns like context degradation, tool confusion, and instruction drift in multi-turn workflows. This surfaces specific breakdowns in interleaved thinking and tool usage to pinpoint why the agent failed.

Interleaved thinking is when an AI agent reasons between tool usages across multiple turns. It causes problems when the reasoning trace degrades, leading to patterns like hallucination, circular reasoning, or goal abandonment that derail the agent workflow.

Optimize prompts by running an analysis loop on captured agent reasoning traces. This detects specific failure patterns and generates improved prompts that correct issues like instruction drift, missing validation, and incomplete reasoning.

Yes, optimization learnings can be converted into reusable Agent Skills. By analyzing reasoning traces and detecting failure patterns, you generate shareable skills that enable team collaboration and prevent repeated agent debugging issues.

Common failure patterns in AI agent reasoning traces include context degradation, tool confusion, instruction drift, hallucination, incomplete reasoning, tool misuse, goal abandonment, circular reasoning, premature conclusion, and missing validation.

No specific dependencies are required to run the reasoning trace optimizer. It operates as a standalone skill to analyze agent workflows, requiring only the captured reasoning traces from your multi-turn AI agent interactions.