Ai Research Explore

This is the skill you reach for when you're already deep in a research project and want to explore candidate improvements without destroying your baseline. It expects you've frozen your task, dataset, and evaluation method, then runs a two-loop rhythm: outer loop gates ideas and decides what's worth trying, inner loop makes one bounded change and collects evidence. Everything lands in explore_outputs with a scientific changelog and comparability report so you know exactly what changed and whether the gain means anything. It won't claim novelty or promise reproduction success, but it will give you auditable candidate ranking against your current anchor. Use it for systematic exploration once you know what you're comparing against, not for open-ended direction finding.

Install to Claude Code

npx -y skills add lllllllama/rigorpilot-skills --skill ai-research-explore --agent claude-code