trace-analysis

To analyze historical learning data for AI agent failure patterns, run the script-based trace analyzer against your project's learning logs to detect and classify recurring issues for targeted skill improvement.

AI agent performance failure patterns are classified into specific categories such as loop, drift, and oversight, allowing you to pinpoint exact behavioral issues and make targeted adjustments to enhance future tasks.

Run a trace analysis on current project learnings by executing the bash trace-analyzer script with your project root directory as the argument to process the historical learning data and output failure recommendations.

This trace analysis skill specifically parses the .agent-harness/learnings.jsonl file format to detect performance failure patterns, so JSONL formatted historical learning logs are required for proper analysis and classification.

The best way to improve AI agent performance after identifying failure patterns is to apply the skill improvement recommendations generated by the trace analysis, which suggest targeted actions based on classified historical failures.

Your AI agent may be experiencing recurring failure patterns like loop or drift. Review its history by running a trace analysis on the learnings JSONL logs to classify failures and receive targeted skill improvement recommendations.