How do I trigger an experiment retrospective?

You can initiate the process by starting your message with the tag or by asking Claude for a summary of lessons learned from your experiment logs.

Does it overwrite my existing documentation automatically?

No, the skill follows a strict 'ask before writing' protocol, presenting proposed changes to your skills and troubleshooting files for your approval first.

Can I use this for non-ML projects?

Yes. While designed with ML experiments in mind, it is effective for any project that involves systematic testing, logging results, and maintaining a lessons-learned database.

What files does this skill specifically monitor?

It identifies relevant data within 'training_reports/*.md', 'references/experiment-log.md', and 'references/troubleshooting.md' to extract findings.

What is a 'Result Skill' in this context?

A Result Skill is a specific Claude instruction file that encapsulates the best practices and recommended configurations discovered during your experiments.

Experiment Retrospective & Insight Extractor

Name: Experiment Retrospective & Insight Extractor
Author: Erland366

byErland366

0•

Data Science & ML

Distills experiment reports and logs into actionable project skills and troubleshooting documentation.

The Experiment Retrospective skill automates the process of analyzing post-experiment data to ensure every lesson learned is captured and applied. By scanning training reports and experiment logs, it extracts critical metrics, hyperparameter results, and failure modes to generate structured summaries. It intelligently updates troubleshooting guides and creates or refines result-based skills, transforming raw data into permanent institutional knowledge. This skill is essential for developers and data scientists who want to bridge the gap between running experiments and building robust, documented workflows.

Key Features

010 GitHub stars

02User-confirmed modification workflow for all repository changes

03Structured logging of historical retrospectives for project tracking

04Extraction of hyperparameters, metrics, and success/failure patterns

05Automated summarization of experiment reports and markdown logs

06Dynamic updates to troubleshooting documentation and skill templates

Use Cases

01Converting experiment failures into permanent troubleshooting entries

02Generating behavioral skills for Claude based on successful experimental outcomes

03Summarizing performance across multiple machine learning training runs

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add erland366/kobo_weixin_translated experiment-retrospective

For use in Claude.ai and ChatGPT

Download Skill