Orchestra Research open-sources a 98-skill AI research library, letting coding agents autonomously run everything from topic selection to paper writing

The library organizes 98 skills across 23 categories, covering research orchestration, topic selection, paper writing, model architecture, fine-tuning, post-training, distributed training, optimization, inference, evaluation, safety alignment, agents, RAG, multimodal, and more. The autoresearch skill acts as a central orchestration layer using a dual-loop architecture of inner optimization plus outer synthesis, routing tasks to domain skills and supporting Claude Code's /loop and OpenClaw heartbeat for continuous operation. Documentation comes from official repos, real GitHub issues, and production workflows. For installation, human users can run npx @orchestra-research/ai-research-skills for an interactive installer that auto-detects installed coding agents (Claude Code, Hermes Agent, OpenCode, Qoder, Cursor, Gemini CLI, etc.), installs skills to ~/.orchestra/skills/ with symlinks, and falls back to copying on Windows. For teams doing post-training and inference deployment, these skills turn key configurations and troubleshooting experience from TRL, GRPO, OpenRLHF, verl, vLLM, SGLang, and other frameworks into agent-readable skill files, effectively baking senior engineers' hard-won lessons into the workflow and reducing time spent debugging infrastructure.