Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
These skill files turn an agent into a research assistant that can design experiments, train models, interpret internals, audit safety, and deploy inference. They are built for AI research scientists, LLM fine-tuning engineers, distributed training engineers, and MLOps teams who need reproducible, agent-guided workflows across the full model lifecycle.
WHO IT'S FOR
AI Research Scientists
orchestrating autonomous research experiments
LLM Fine-Tuning Engineers
fine-tuning models with RL and preference optimization
Distributed Training Engineers
scaling LLM training across GPU clusters
MLOps and Infrastructure Teams
managing experiment tracking and GPU infrastructure
Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.
Follow the documented setup, then try a first task.
I’m using
Interactive AI Research Skills installer
Installs: AI Research Skills library: choose everything, the quickstart bundle, categories, or individual skills. Skills are installed to ~/.orchestra/skills/ and linked to detected agents. · Claude Code, Hermes Agent, OpenCode, Qoder, Cursor, Gemini CLI
Before you start
Platform: Uses symlinks, with a copy fallback on Windows.
Run this command in a terminal to open the interactive installer.
npx @orchestra-research/ai-research-skills
Choose the detected coding agents and the desired selection: everything, quickstart bundle, categories, or individual skills.
No dataset or running training environment is needed for this configuration walkthrough.
Make sure axolotl is installed in your agent before trying this task.
Use the axolotl skill to show the documented FSDP2 YAML configuration for a Llama model and explain its parameter offloading, state dictionary, layer wrapping, and resharding settings. Keep this to a configuration walkthrough without starting training.
Python environment with faiss-cpu and NumPy available if you want to run the example.
No external dataset is required; the example generates sample vectors.
Make sure faiss is installed in your agent before trying this task.
Use the faiss skill to write a minimal Python example that creates 1,000 random float32 vectors with 128 dimensions, adds them to an IndexFlatL2 index, and searches for the five nearest neighbors of one random query. Explain the returned indices and distances.
An existing SentencePiece tokenizer model; replace the sample path m.model with its actual path.
Python environment with sentencepiece installed if you want to run the example.
Make sure sentencepiece is installed in your agent before trying this task.
Use the sentencepiece skill to write a short Python example that loads my existing tokenizer model, encodes 'This is a test' into both pieces and token IDs, and decodes the IDs back to text. Explain the whitespace marker in the pieces. Treat m.model as a placeholder for my actual model path.
Claude
Codex
Cursor
Gemini CLI
OpenClaw
OpenCode
Hermes
Qwen Code
DeepSeek