#!/bin/bash
# analyze-paper.sh - Extract structured information from research papers
# Version: 1.0.0
# Part of: Paper2Agent Skill
set -euo pipefail
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
CACHE_DIR="${MIYABI_PAPER2AGENT_CACHE_DIR:-$HOME/.miyabi/paper2agent}"
PDF_PARSER="${MIYABI_PAPER2AGENT_PDF_PARSER:-pdftotext}"
mkdir -p "$CACHE_DIR"
# ============================================================
# Functions
# ============================================================
log() {
echo "[$(date -Iseconds)] $*" >&2
}
download_paper() {
local url="$1"
local output="$2"
log "Downloading paper from $url"
if [[ "$url" =~ ^https://arxiv.org/abs/ ]]; then
# Convert abs URL to PDF URL
url="${url/\/abs\//\/pdf\/}.pdf"
fi
curl -L -o "$output" "$url"
log "Downloaded to $output"
}
extract_text() {
local pdf_file="$1"
local txt_file="$2"
log "Extracting text from PDF"
case "$PDF_PARSER" in
pdftotext)
if ! command -v pdftotext &>/dev/null; then
log "ERROR: pdftotext not found. Install: brew install poppler"
exit 1
fi
pdftotext -layout "$pdf_file" "$txt_file"
;;
pypdf)
python3 <<EOF
from pypdf import PdfReader
reader = PdfReader("$pdf_file")
with open("$txt_file", "w") as f:
for page in reader.pages:
f.write(page.extract_text())
EOF
;;
*)
log "ERROR: Unknown PDF parser: $PDF_PARSER"
exit 1
;;
esac
log "Text extracted to $txt_file"
}
extract_section() {
local text_file="$1"
local section_name="$2"
# Try different section header formats
local patterns=(
"^${section_name}\$"
"^${section_name}[[:space:]]*\$"
"^[0-9]+\.?[[:space:]]*${section_name}"
"^${section_name}:"
)
for pattern in "${patterns[@]}"; do
local result=$(awk -v pattern="$pattern" '
BEGIN { capturing = 0; }
$0 ~ pattern { capturing = 1; next; }
capturing && /^[A-Z][A-Za-z ]+$/ && !/^(The|A|An|In|For|To|Of)/ { exit; }
capturing { print; }
' "$text_file")
if [ -n "$result" ]; then
echo "$result"
return 0
fi
done
return 1
}
extract_github_urls() {
local text_file="$1"
grep -Eo "https://github\.com/[a-zA-Z0-9_-]+/[a-zA-Z0-9_-]+" "$text_file" | sort -u
}
extract_arxiv_id() {
local text_file="$1"
# Try to find arXiv ID in various formats
grep -Eo "arXiv:[0-9]{4}\.[0-9]{4,5}" "$text_file" | head -1 | sed 's/arXiv://'
}
extract_title() {
local text_file="$1"
# Take first few lines and find the longest line (likely title)
head -20 "$text_file" | awk '
length > max_length { max_length = length; longest = $0; }
END { print longest; }
' | sed 's/^[[:space:]]*//;s/[[:space:]]*$//'
}
generate_json_output() {
local title="$1"
local abstract="$2"
local methods="$3"
local github_urls="$4"
local arxiv_id="$5"
# Escape JSON strings
title=$(echo "$title" | jq -R .)
abstract=$(echo "$abstract" | jq -Rs .)
methods=$(echo "$methods" | jq -Rs .)
# Convert GitHub URLs to JSON array
local github_json="[]"
if [ -n "$github_urls" ]; then
github_json=$(echo "$github_urls" | jq -R . | jq -s .)
fi
cat <<EOF
{
"title": $title,
"arxiv_id": "$(echo "$arxiv_id" | jq -R .)",
"abstract": $abstract,
"methods": $methods,
"code_references": $github_json,
"extracted_at": "$(date -Iseconds)",
"parser": "$PDF_PARSER"
}
EOF
}
# ============================================================
# Main Logic
# ============================================================
usage() {
cat <<EOF
Usage: $0 <paper-url-or-path> [output-json]
Extract structured information from a research paper PDF.
Arguments:
paper-url-or-path URL (arXiv, etc.) or local file path
output-json Output JSON file (default: stdout)
Environment:
MIYABI_PAPER2AGENT_CACHE_DIR Cache directory (default: ~/.miyabi/paper2agent)
MIYABI_PAPER2AGENT_PDF_PARSER PDF parser: pdftotext (default) or pypdf
Examples:
# From arXiv URL
$0 https://arxiv.org/abs/2509.06917 > analysis.json
# From local PDF
$0 ./paper.pdf analysis.json
# With pypdf parser
MIYABI_PAPER2AGENT_PDF_PARSER=pypdf $0 paper.pdf
EOF
exit 1
}
main() {
if [ $# -lt 1 ]; then
usage
fi
local input="$1"
local output_json="${2:--}" # Default to stdout
local pdf_file
local is_temp=false
# Determine if input is URL or file
if [[ "$input" =~ ^https?:// ]]; then
# URL - download to cache
local cache_name=$(echo "$input" | md5sum | cut -d' ' -f1)
pdf_file="$CACHE_DIR/$cache_name.pdf"
if [ ! -f "$pdf_file" ]; then
download_paper "$input" "$pdf_file"
else
log "Using cached PDF: $pdf_file"
fi
else
# Local file
if [ ! -f "$input" ]; then
log "ERROR: File not found: $input"
exit 1
fi
pdf_file="$input"
fi
# Extract text
local text_file="$CACHE_DIR/$(basename "$pdf_file" .pdf).txt"
extract_text "$pdf_file" "$text_file"
# Extract sections
log "Extracting sections"
local title=$(extract_title "$text_file")
log "Title: $title"
local abstract=$(extract_section "$text_file" "Abstract" || echo "")
log "Abstract: ${#abstract} chars"
local methods=$(extract_section "$text_file" "Methods" || \
extract_section "$text_file" "Methodology" || \
extract_section "$text_file" "Approach" || echo "")
log "Methods: ${#methods} chars"
# Extract metadata
local github_urls=$(extract_github_urls "$text_file" || echo "")
local github_count=$(echo "$github_urls" | wc -l | tr -d ' ')
log "GitHub URLs: $github_count found"
local arxiv_id=$(extract_arxiv_id "$text_file" || echo "")
log "arXiv ID: $arxiv_id"
# Generate JSON output
local json_output=$(generate_json_output "$title" "$abstract" "$methods" "$github_urls" "$arxiv_id")
if [ "$output_json" = "-" ]; then
echo "$json_output"
else
echo "$json_output" > "$output_json"
log "Analysis saved to $output_json"
fi
}
# ============================================================
# Entry Point
# ============================================================
if [ "${BASH_SOURCE[0]}" == "${0}" ]; then
main "$@"
fi
