Copy repositories to make them easily digestable for LLMs.

llmfeed.sh

#!/bin/bash
# llmfeed.sh - Copy repository contents to make them easily digestable for LLMs
#
# Author: Felix Weiglhofer
# License: MIT
#
# Description:
# Flattens the (text) content of a repository into a single string that can be
# easily read by LLMs. Inspired by uithub.com.
#
# Usage:
#   - Copy to clipboard (Wayland):
#       ./llmfeed.sh <targetDir> | wl-copy
#
#   - Copy to clipboard (X11):
#       ./llmfeed.sh <targetDir> | xclip -selection clipboard
#
# TODO:
#   - Add flag to estimate token count
#   - Add flag to prune first N lines from file (remove license header to save tokens)
#

targetDir=$1
if [[ - z $targetDir ]]; then
    echo "Error: Target directory missing."
    echo ""
    echo "Usage: $0 <targetDir>"
    exit 1
fi

findFlags=(
    # Exclude directories
    -path '*/.git'  -prune -o
    -path '*/build' -prune -o

    -type f -print0
)

is_binary() {
    if [[ -z $(file -b --mime-type "$1" | grep -v "^text/") ]]; then
        return 1
    else
        return 0
    fi
}

mapfile -d '' -t listOfFile < <(find $targetDir ${findFlags[@]})

content=""
for file in "${listOfFile[@]}"
do
    if is_binary $file; then
        continue
    fi
    content+=">>> $file"
    content+=$'\n'
    content+=$(cat $file)
    content+=$'\n'
done

echo "$content"