Script to take a remote git repo (e.g. GitHub) and create some basic documentation for it

make_docs.sh

#!/bin/bash

# Check if a repository URL was provided
if [ $# -ne 1 ]; then
    echo "Usage: $0 <repository-url>"
    exit 1
fi

REPO_URL=$1

# Extract repo name from URL
# This will get the last part of the URL without .git extension
REPO_NAME=$(basename "$REPO_URL" .git)

# Create a temporary directory for our work
TEMP_DIR=$(mktemp -d)
FILE_LIST="$TEMP_DIR/file_list.txt"

# Clone the repository
git clone "$REPO_URL" "$TEMP_DIR/project"

cd $TEMP_DIR

# Generate the file list using Claude
echo "Given a list of directories and files in a software repo, list which ones would be useful to scan to put together some documentation? Think tests and README style documents. Only return a list of the matching paths. Put the README first, if possible. Here are the paths:" | cat - <(find project -maxdepth 1 -mindepth 1) | llm -m claude-3.5-sonnet -s "Return only a list of paths with no preamble or extra explanation" > "$FILE_LIST"

# Generate the documentation
cat "$FILE_LIST" | grep -v '^\.|\/' | tr '\n' ' ' | xargs files-to-prompt -c > "$TEMP_DIR/prompt.txt"

cd -

cat "$TEMP_DIR/prompt.txt" | ttok -t 150000 > "$TEMP_DIR/prompt2.txt"

cat "$TEMP_DIR/prompt2.txt" | llm -m claude-3.5-sonnet -s 'write detailed usage documentation for this project including realistic examples. do not include a preamble, go straight into the documentation' > "${REPO_NAME}.md"

# Clean up
rm -f "$FILE_LIST"
#rm -rf "$TEMP_DIR"

echo "Documentation generated in ${REPO_NAME}.md"

This works but does have a flaw in that passing complex temporary folder names to LLMs can... be problematic. So I want to clean that up first before making this public.

Cheated a bit by changing directory and back again. Also added token truncation with ttok.

I want to come up with a safer way of handling the cleaning up at the end, because rm -rf anything is terrifying in any case.

Apparently macOS, at least, cleans up any mktemp directories on reboot, so in theory I could ditch the clean up.

Commented it out for now, will think about it.