This repository implements the PaperOrchestra multi-agent pipeline as a set of host-agent-executable skills. Each skill consists of instruction documents (SKILL.md) and reference materials that guide a coding agent through a five-stage workflow: outline generation, plotting, literature review, section writing, and content refinement. There are no embedded LLM calls or API keys — the host agent (Claude Code, Cursor, Cline, etc.) performs all reasoning and web searches using its own tools. The system is benchmark-driven with autoraters for evaluation.