Concatenate a directory full of files into a single prompt for use with LLMs
https://github.com/simonw/files-to-prompt.git
Concatenate a directory full of files into a single prompt for use with LLMs
For background on this project see Building files-to-prompt entirely using Claude 3 Opus.
Install this tool using pip:
pip install files-to-prompt
To use files-to-prompt, provide the path to one or more files or directories you want to process:
files-to-prompt path/to/file_or_directory [path/to/another/file_or_directory ...]
This will output the contents of every file, with each file preceded by its relative path and separated by ---.
-e/--extension <extension>: Only include files with the specified extension. Can be used multiple times.files-to-prompt path/to/directory -e txt -e md
--include-hidden: Include files and folders starting with . (hidden files and directories).files-to-prompt path/to/directory --include-hidden
--ignore <pattern>: Specify one or more patterns to ignore. Can be used multiple times. Patterns may match file names and directory names, unless you also specify --ignore-files-only. Pattern syntax uses fnmatch, which supports *, ?, [anychar], [!notchars] and [?] for special character literals.files-to-prompt path/to/directory --ignore "*.log" --ignore "temp*"
--ignore-files-only: Include directory paths which would otherwise be ignored by an --ignore pattern.files-to-prompt path/to/directory --ignore-files-only --ignore "*dir*"
--ignore-gitignore: Ignore .gitignore files and include all files.files-to-prompt path/to/directory --ignore-gitignore
-c/--cxml: Output in Claude XML format.files-to-prompt path/to/directory --cxml
-m/--markdown: Output as Markdown with fenced code blocks.files-to-prompt path/to/directory --markdown
-o/--output <file>: Write the output to a file instead of printing it to the console.files-to-prompt path/to/directory -o output.txt
-n/--line-numbers: Include line numbers in the output.files-to-prompt path/to/directory -n
Example output:
files_to_prompt/cli.py
---
1 import os
2 from fnmatch import fnmatch
3
4 import click
...
-0/--null: Use NUL character as separator when reading paths from stdin. Useful when filenames may contain spaces.find . -name "*.py" -print0 | files-to-prompt --null
Suppose you have a directory structure like this:
my_directory/
โโโ file1.txt
โโโ file2.txt
โโโ .hidden_file.txt
โโโ temp.log
โโโ subdirectory/
โโโ file3.txt
Running files-to-prompt my_directory will output:
my_directory/file1.txt
---
Contents of file1.txt
---
my_directory/file2.txt
---
Contents of file2.txt
---
my_directory/subdirectory/file3.txt
---
Contents of file3.txt
---
If you run files-to-prompt my_directory --include-hidden, the output will also include .hidden_file.txt:
my_directory/.hidden_file.txt
---
Contents of .hidden_file.txt
---
...
If you run files-to-prompt my_directory --ignore "*.log", the output will exclude temp.log:
my_directory/file1.txt
---
Contents of file1.txt
---
my_directory/file2.txt
---
Contents of file2.txt
---
my_directory/subdirectory/file3.txt
---
Contents of file3.txt
---
If you run files-to-prompt my_directory --ignore "sub*", the output will exclude all files in subdirectory/ (unless you also specify --ignore-files-only):
my_directory/file1.txt
---
Contents of file1.txt
---
my_directory/file2.txt
---
Contents of file2.txt
---
The tool can also read paths from standard input. This can be used to pipe in the output of another command:
# Find files modified in the last day
find . -mtime -1 | files-to-prompt
When using the --null (or -0) option, paths are expected to be NUL-separated (useful when dealing with filenames containing spaces):
find . -name "*.txt" -print0 | files-to-prompt --null
You can mix and match paths from command line arguments and stdin:
# Include files modified in the last day, and also include README.md
find . -mtime -1 | files-to-prompt README.md
Anthropic has provided specific guidelines for optimally structuring prompts to take advantage of Claude's extended context window.
To structure the output in this way, use the optional --cxml flag, which will produce output like this:
<documents>
<document index="1">
<source>my_directory/file1.txt</source>
<document_content>
Contents of file1.txt
</document_content>
</document>
<document index="2">
<source>my_directory/file2.txt</source>
<document_content>
Contents of file2.txt
</document_content>
</document>
</documents>
The --markdown option will output the files as fenced code blocks, which can be useful for pasting into Markdown documents.
files-to-prompt path/to/directory --markdown
The language tag will be guessed based on the filename.
If the code itself contains triple backticks the wrapper around it will use one additional backtick.
Example output:
`%%CODEBLOCK23%%python
def my_function():
return "Hello, world!"
%%CODEBLOCK24%%javascript
function myFunction() {
return "Hello, world!";
}
%%CODEBLOCK25%%markdown
This file has its own
fenced code blocks
Inside it.
%%CODEBLOCK27%%`
To contribute to this tool, first checkout the code. Then create a new virtual environment:
cd files-to-prompt
python -m venv venv
source venv/bin/activate
Now install the dependencies and test dependencies:
pip install -e '.[test]'
To run the tests:
pytest