What files does the skill analyze to provide advice?

It scans .codex/skills/ for relevant results, training_reports/*.md for recent performance, experiment-log.md for history, and troubleshooting.md for error patterns.

How do I trigger the Experiment Advisor skill?

You can trigger it by starting your message with the tag or by asking 'What should we try next?' in the context of an experiment.

Can the advisor handle hardware or budget constraints?

Yes, you can specify hardware limitations or run budgets in your goal, and the advisor will tailor its proposed experiment plan accordingly.

How does it help avoid previous training errors?

The skill specifically checks references/troubleshooting.md to identify failure patterns and explicitly designs the new plan to avoid those pitfalls.

Does it automatically run the experiments?

No, it provides a structured markdown table with 2–5 proposed plans (including batch size, learning rates, etc.) for you to review and implement.

Experiment Advisor

Name: Experiment Advisor
Author: Erland366

byErland366

0•

Data Science & ML

Proposes structured, data-driven experiment plans by analyzing historical training reports, logs, and research goals.

Experiment Advisor is designed for researchers and machine learning engineers who need guidance on the next steps of their development cycle. By scanning local training reports, experiment logs, and troubleshooting documentation, the skill generates 2–5 concrete experiment proposals tailored to your hardware and budget constraints. It ensures continuity in research by leveraging successful defaults from existing skills while explicitly avoiding documented failure patterns to optimize the iteration process.

Key Features

01Integrates with troubleshooting documentation to avoid known errors

020 GitHub stars

03Analyzes historical training reports and experiment logs for context

04Optimizes plans based on hardware constraints and run budgets

05Generates 2–5 structured experiment proposals with specific hyperparameters

06Appends summarized plans to the experiment log for team visibility

Use Cases

01Designing data mixture experiments for complex machine learning tasks

02Deciding on hyperparameter adjustments after reviewing recent training logs

03Planning model architecture iterations based on documented performance benchmarks

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add erland366/kobo_weixin_translated experiment-advisor

For use in Claude.ai and ChatGPT

Download Skill