Which technical domains are currently supported?

The skill features specialized logic for Research (ML training), Unsloth (efficient LLM fine-tuning), and CUDA (GPU kernel development).

How do I trigger the Domain Advisor skill?

You can activate it by starting a message with the tag or asking Claude planning questions like 'What should we try next?'

How does Domain Advisor know what to suggest?

It analyzes your project's registry.json, past experiment reports, and logs to identify trends, reuse successful defaults, and avoid previous errors.

Can it help me avoid recurring bugs or failures?

Yes, it specifically checks your project's troubleshooting guides to ensure proposed plans do not repeat known failure patterns.

Does it automatically update my experiment logs?

It can append short summaries of proposed plans to your experiment log, but only after receiving explicit user approval.

Domain Advisor

Name: Domain Advisor
Author: Erland366

byErland366

データサイエンスとML

Recommends optimized experiment plans and development tasks by analyzing historical project data across machine learning and GPU computing domains.

概要

The Domain Advisor skill transforms Claude into a strategic planning assistant for specialized technical projects. By scanning project registries, experiment logs, and troubleshooting guides, it provides context-aware suggestions for next steps in research, LLM fine-tuning via Unsloth, or CUDA kernel optimization. It helps engineers avoid past failures and maximize efficiency by proposing concrete experiments with clear objectives, expected outcomes, and configuration parameters tailored to the specific domain's constraints and historical performance.

主な機能

0 GitHub stars
Context-aware experiment planning based on historical project logs and reports
Domain-specific logic for Research, Unsloth fine-tuning, and CUDA development
Automated retrieval of troubleshooting guides to bypass known failure patterns
Seamless integration with project registries for path and resource management
Standardized experiment tables featuring objectives, parameters, and hypotheses

ユースケース

Determining the next set of hyperparameters to test in a machine learning training sweep
Optimizing LoRA configurations and memory settings for efficient LLM fine-tuning
Planning tiling and shared memory strategies for high-performance CUDA kernel optimization

概要

主な機能

0 GitHub stars
Context-aware experiment planning based on historical project logs and reports
Domain-specific logic for Research, Unsloth fine-tuning, and CUDA development
Automated retrieval of troubleshooting guides to bypass known failure patterns
Seamless integration with project registries for path and resource management
Standardized experiment tables featuring objectives, parameters, and hypotheses

ユースケース

Determining the next set of hyperparameters to test in a machine learning training sweep
Optimizing LoRA configurations and memory settings for efficient LLM fine-tuning
Planning tiling and shared memory strategies for high-performance CUDA kernel optimization