flewor2

AI Research Scientist Skill — AI skill for Claude Code

AI community

Claude Code skill: work like a 2026 AI research scientist — research discipline, validation, SOTA recipes (LLM pre/post-training, diffusion & flow matching, CV, NLP) and a Kaggle winning-methods surve.

How to install AI Research Scientist Skill

This entry records only its repository, not the path inside it, so there is no exact command to give. Open flewor2/ai-research-scientist-skill and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AI Research Scientist Skill does

Claude Code skill: work like a 2026 AI research scientist — research discipline, validation, SOTA recipes (LLM pre/post-training, diffusion & flow matching, CV, NLP) and a Kaggle winning-methods survey

Alternatives in AI

  • Claude Cookbook — Official notebooks and recipes for common patterns (RAG, tool use, Skills, MCP) 35.3k ★
  • Post Training — ai-research-skills GRPO, RLHF, DPO, SimPO 5.4k ★
  • ML Training Pack — Machine learning model training skills 1.6k ★

README

ai-research-scientist

A Claude Code skill that makes Claude work like a strong 2026 AI research scientist and research engineer.

The skill puts research discipline first:

  • a hello-world sanity check before any experiment,
  • validation design (k-fold variants, leakage rules, adversarial validation),
  • strong and equally tuned baselines,
  • error bars and paired tests,
  • reproducible code.

The reference files then give state-of-the-art recipes (snapshot of 2026-10-04).

Contents

Path Content
SKILL.md Operating principles, a routing table by request type, the standard experiment workflow
references/methodology.md Research loop, experiment design, cross-validation, evals and the 2026 benchmark status, statistics, paper writing, red flags
references/engineering-practices.md Project layout, configs, reproducibility, testing ML code, logging, checkpoints, silent-bug catalog, release checklist
references/pretraining.md Frontier LLM reports, architecture, MoE, Muon, schedules, scaling laws, data, FP8/NVFP4, default recipes
references/posttraining.md SFT, DPO, RLVR (GRPO/DAPO/GSPO/CISPO), async RL, reward models, agentic RL, RL debugging
references/diffusion-flow-matching.md Diffusion and flow-matching theory and practice, guidance, DiT/MMDiT, few-step methods, diffusion LMs, FID hygiene, minimal tested code
references/computer-vision.md Backbones, fine-tuning recipe, detection, segmentation, depth/3D, OCR, deployment, CV silent bugs
references/nlp.md Encoders, embeddings and rerankers, RAG, extraction, small-LLM LoRA, NLP silent bugs
references/systems-and-modalities.md Compute/memory math, parallelism, kernels, inference, VLMs, SSL, robotics, science AI
references/kaggle-winning-methods.md Survey of ~98 Kaggle competitions (2023–2026): leaderboards, top solutions, method frequency, prioritized playbook
scripts/sanity_check.py Hello world: environment and GPU check,