| name | paper-to-skill |
| description | Use when extracting reusable AI4Math research skills from papers, especially PDF-to-SkillCard workflows, paper triage, PDF preparation, proof-pattern extraction, and cross-paper SkillCard synthesis for a coding-agent-neutral paper-to-skill workflow. |
Paper To Skill
This repository is a Skill-first paper-to-skill workflow package for AI4Math auto research. Codex is the reference operator, but the shared Skill layer is intended for any capable coding agent.
Entry Point
Start user-facing paper-to-skill work from:
skills/paper-to-skill-workflow/SKILL.md
Use skills/registry.yaml to inspect phase order, review gates, and optional upstream or downstream Skills.
This root SKILL.md is not the workflow driver. It is a compatibility entrypoint for platforms that expect one top-level Skill file.
Operating Boundary
- Accept
paper.pdf as the default user input; accept paper.md when conversion has already happened.
- Preserve the original
paper.pdf when provided and the full converted paper.md.
- Do not require external LLM APIs for extraction; the active coding agent is the extraction engine.
- Keep platform-specific adapters thin and point them back to
skills/registry.yaml and the shared skills/ layer.
- Every accepted or review-worthy SkillCard must include
source.paper_md, source.start_line, and source.end_line.
- Treat structured JSON, YAML, and reports as indexes or review artifacts, not replacements for the original paper.
- Use human feedback and review decisions as explicit workflow state when available.