gen-report-concise

What problem does it solve?

This Skill automates creation of compact, reproducible experiment reports by extracting key metrics, configuration, and provenance from training run artifacts so engineers and researchers can quickly understand results and trade-offs without manual log parsing.

Core Features & Use Cases

• Automated data collection: Parallel read of training logs, snapshot model configs, training toml files, CSV metrics, and artifacts to gather numeric and contextual data.
• Provenance-first metrics: Every reported metric includes mathematical definition, physical meaning, and exact data source with file path, line number, and step identifier.
• Structured concise output: Fills a four-section report template that includes takeaway, motivation, reproducibility, and results with a final trade-off table.
• Use case: Summarize flame_moonshot training runs for quick handoff, postmortem, or performance comparison across experiments.

Quick Start

Generate a concise experiment report for experiment kda_680M in /nfs/ridgerzhu/flame_moonshot/exp/ by invoking the gen-report-concise skill and include logs, configs, and metrics for provenance.