manage-runs

To inspect pipeline runs, scan run directories under working/runs and parse pipeline_state.json to extract per-run status, timing, and agent counts for quick debugging summaries.

Yes, you can compare two pipeline runs side-by-side to view differences in agent progress, timing, and outputs, which directly supports identifying bottlenecks and optimization opportunities.

The best way to clean up old pipeline runs is to scan the working/runs directory and remove run folders older than 30 days, with user confirmation, while preserving recent outputs and state files.

Pipeline run scanning handles missing or corrupted pipeline_state.json files by applying edge-case logic that delivers graceful messaging instead of failing, ensuring legacy formats and unknown statuses are reported clearly.

You need self-contained run directories named by date or dataset under working/runs, each containing its own working files, outputs, and pipeline_state.json, to successfully browse and extract run artifacts.

A pipeline run summary includes extracted pipeline_state.json fields such as per-run status, timing data, agent counts, output artifacts, and validation confidence metrics when available.