autosearch:trace-harvest

To extract reusable patterns from session traces, you run trace-harvest on completed archives to parse tool calls, branches, and recoveries into structured, compact outputs. This distills noisy data into promote-candidate patterns for knowledge retention.

Session trace analysis for experience capture is the process of distilling successful session tool-call data into compact, reusable patterns. It parses inputs, tool choices, branches, and recoveries to identify effective paths and store provenance with a trace_ref.

Batch processing session traces nightly involves parsing trace structures including inputs, tool calls, branches, and recoveries to identify effective paths. The process emits per-skill patterns.jsonl files to keep experience memory up to date.

You can use trace-harvest on completed Autosearch sessions to generate winning patterns for experience capture. It analyzes complete traces including inputs, tool choices, branches, and recoveries to identify effective paths and produce promote-candidate patterns.

The best way to store winning patterns for knowledge retention is to emit them into per-skill patterns.jsonl files. This skill structures the extracted patterns with a trace_ref for provenance, feeding downstream learning systems.