paper-to-skill-workflow

name	paper-to-skill-workflow
description	Run the complete AI4Math paper-to-skill workflow from a user-provided PDF or Markdown paper. Use when a coding agent should preserve the original paper, convert PDF to paper.md when needed, extract proof-pattern SkillCandidates and SkillCards with line evidence, and optionally route reviewed cards into cross-paper synthesis.

Paper To Skill Workflow

Purpose

Use this skill as the default user-facing entrypoint. A user should be able to provide a paper.pdf and an output directory, then receive reusable proof_pattern SkillCandidates, SkillCards, and a review report.

If the user already provides paper.md, skip conversion and run extraction directly. Still keep the workflow-level artifact structure.

Inputs

Accept:

• paper.pdf as the primary user input.
• paper.md when the paper has already been converted.
• Optional metadata.json.
• Optional human_feedback_state.json.
• Optional output directory such as outputs/<research_project_id>/papers/<paper_id>/.

Default Flow

paper.pdf
  -> pdf-to-markdown-converter
  -> paper.md
  -> paper-to-skill-extractor
  -> skill_candidates.json + skill_cards/*.yaml + report.md

For already-converted papers:

paper.md
  -> paper-to-skill-extractor
  -> skill_candidates.json + skill_cards/*.yaml + report.md

Workflow

1. Choose or create an output paper directory. Prefer outputs/<research_project_id>/papers/<paper_id>/ when the user is running a project, otherwise use a clear local output folder derived from the input filename.
2. Preserve the original paper.pdf when a PDF is provided.
3. If the input is PDF, use pdf-to-markdown-converter to create paper.md, mineru/, and conversion_report.json.
4. Treat the full paper.md as the source for line-referenced extraction evidence.
5. If human_feedback_state.json exists, pass it through to the extraction step.
6. Use paper-to-skill-extractor to produce skill_candidates.json, skill_cards/*.yaml, and report.md.
7. In the final response, tell the user which artifacts were written and whether any candidates need review.
8. If the user asks to synthesize across papers, route accepted or review-worthy SkillCards to cross-paper-skill-synthesizer.

Review Gates

• Ask before downloading PDFs from the network.
• Ask before installing or changing dependencies for PDF conversion.
• Ask before accepting unsafe mathematical generalizations.
• Ask before merging or generalizing SkillCards across papers.

Completion Check

Before finishing, confirm:

• The original paper.pdf is preserved when provided.
• The full paper.md exists.
• skill_candidates.json is valid JSON.
• Each accepted or review-worthy SkillCard has source.paper_md, source.start_line, and source.end_line.
• report.md separates accepted, needs-review, and rejected candidates.