agent-survey-corpus_skill

Overview

This skill downloads a small corpus of open-access arXiv survey and review PDFs about LLM agents and extracts text to a local reference folder. It helps you learn how real agent surveys structure sections, allocate subsection sizes, and present evidence-backed comparisons for style and structure modeling.

How this skill works

You supply a plain list of arXiv IDs in ref/agent-surveys/arxiv_ids.txt and run the downloader script. The tool fetches PDFs from arXiv, saves them under ref/agent-surveys/pdfs/, extracts the first N pages to plain text under ref/agent-surveys/text/, and generates a STYLE_REPORT.md summarizing section counts and observed patterns.

When to use it

• When you want concrete examples of how agent survey papers organize 6–8 H2 sections and H3 subsections.
• When refining an outline or writing pipeline components that must mimic paper-like structure and citation-backed claims.
• Before drafting comparative tables and claims to ensure your style follows existing community norms.
• When building a small, local training/analysis corpus without relying on external indices.
• When you need extractable text snippets for style learning or rhetorical pattern identification.

Best practices

• Keep only arXiv IDs in ref/agent-surveys/arxiv_ids.txt, one per line.
• Limit max-pages to the sections you actually need (default 20) to save bandwidth and disk space.
• Use --sleep to avoid transient download issues and respect arXiv rate limits.
• Store raw PDFs and extracted text under ref/ and keep these paths git-ignored to avoid large commits.
• Rerun with --overwrite when you update IDs or want fresh extracts.

Example use cases

• Survey-writing: analyze sectioning and sentence density to shape a new review paper outline.
• Pipeline tuning: extract rhetorical patterns for automated outline-to-paragraph generators.
• Comparative claims: gather evidence-backed phrasing examples for table captions and claim language.
• Style reports: auto-generate STYLE_REPORT.md as a checklist for desired section counts and transitions.
• Education: teach team members how published agent surveys structure introductions, methods, and evaluations.