Explore Code

This is for when you want Claude to try exploratory changes in a deep learning codebase without touching your trusted baseline. It works on isolated branches to transplant modules, swap backbones, add LoRA layers, or stitch together low-risk migration ideas. Everything goes into `explore_outputs/` with rollback instructions and clear candidate status. It's explicitly not for baseline reproduction, conservative debugging, or anything that needs verification rigor. The boundaries are sharp: you must authorize the exploration, the code stays isolated, and the output is always treated as a candidate. Useful when you want to prototype architectural changes quickly while keeping a clear separation from production research code.

skills.sh installs: 237
GitHub stars: 289
last updated: Jun 3, 2026

View on GitHub

Install

$ npx -y skills add lllllllama/ai-paper-reproduction-skill --skill explore-code --agent claude-code

Installs into .claude/skills of the current project.

Files

explore-code

Use this as the Rigor Improve implementation leaf skill. The installed slug remains explore-code for compatibility.

Use the shared operating principles in ../ai-research-reproduction/references/agent-operating-principles.md; this skill should guide bounded candidate code work without over-prescribing implementation details.

When to apply

When the researcher explicitly authorizes exploratory code changes on an isolated branch or worktree.
When the task is source-anchored module transplant, backbone adaptation, LoRA or adapter insertion, or low-risk module combination.
When summary-level recording is sufficient and the result is a candidate, not a trusted conclusion.

When not to apply

When the request is for trusted baseline work, conservative debugging, or normal training execution.
When the user did not explicitly authorize exploratory modifications.
When the task is a broad refactor or a from-scratch idea implementation.