Simple, readable metadata attached to files can be useful in numerous situations, such as recording title, author, source, copyright, or the provenance of a file.
Unfortunately, it's often unclear how to format such metadata consistently across different file types while preserving valid syntax, making parsing easy, and breaking interoperability with existing tools.
Frontmatter format is a way to add metadata as frontmatter on any file. It is basically a micro-format: a simple set of conventions to put structured metadata as YAML at the top of a file in a syntax that is broadly compatible with programming languages, browsers, editors, and other tools.
Frontmatter format specifies a syntax for the metadata as a comment block at the top of a file. This approach works while ensuring the file remains valid Markdown, HTML, CSS, Python, C/C++, Rust, SQL, or most other text formats.
Frontmatter format is a generalization of the common format for frontmatter used by Jekyll
and other CMSs for Markdown files.
In that format, frontmatter is enclosed in lines containing --- delimiters.
In this generalized format, we allow several styles of frontmatter demarcation, with the first line of the file indicating the format and style.
This repository is a description of the format and an easy-to-use reference implementation. The implementation is in Python but the format is very simple and easy to implement in any language.
This readme aims to explain the format so anyone can use it and encourage the adoption of the format, especially for workflows around text documents that are becoming increasingly common in AI tools and pipelines.
Some simple examples:
---
title: Sample Markdown File
state: draft
created_at: 2022-08-07 00:00:00
tags:
- yaml
- examples
# This is a YAML comment, so ignored.
---
Hello, *World*!<!---
title: Sample HTML File
--->
Hello, <i>World</i>!#---
# author: Jane Doe
# description: A sample Python script
#---
print("Hello, World!")/*---
filename: styles.css
---*/
.hello {
color: green;
}----
-- title: Sample SQL Script
----
SELECT * FROM world;Here's an example of a richer metadata in use, from a tool that does video transcription. You can see how it's useful having a simple and clear format for title, description, history, source of the content, etc.
-
Compatible with existing syntax: By choosing a style for the metadata consistent with any given file, it generally doesn't break existing tools. Almost every language has a style for which frontmatter works as a comment.
-
Auto-detectable format: Frontmatter and its format can be recognized by the first few bytes of the file. That means it's possible to detect metadata and parse it automatically.
-
Metadata is optional: Files with or without metadata can be read with the same tools. So it's easy to roll out metadata into files gracefully, as needed file by file.
-
YAML syntax: JSON, YAML, XML, and TOML are all used for metadata in some situations. YAML is the best choice here because it is already in widespread use with Markdown, is a superset of JSON (in case an application wishes to use pure JSON), and is easy to read and edit manually.
A file is in frontmatter format if the first characters are one of the following:
-
--- -
<!--- -
#--- -
//--- -
/*--- -
-----
and if this prefix is followed by a newline (\n).
The prefix determines the style of the frontmatter. The style specifies the matching terminating delimiter for the end of the frontmatter as well as an optional prefix (which is typically a comment character in some language).
The allowed frontmatter styles are:
-
YAML style: delimiters
---and---with no prefix on each line. Useful for text or Markdown content. -
HTML style: delimiters
<!---and--->with no prefix on each line. Useful for HTML or XML or similar content. -
Hash style: delimiters
#---and#---with#prefix on each line. Useful for Python or similar code content. Also works for CSV files for some (but sadly not all) tools. -
Slash style: delimiters
//---and//---with//prefix on each line. Useful for Rust or C++ or similar code content. -
Slash star style: delimiters
/*---and---*/with no prefix on each line. Useful for JavaScript, TypeScript, CSS or C or similar code content. -
Dash style: delimiters
----and----with--prefix on each line. Useful for SQL or similar code content.
The delimiters must be alone on their own lines, terminated with a newline.
Any style is acceptable on any file as it can be automatically detected. When writing, you can specify the style.
For all frontmatter styles, the content between the delimiters is YAML text in UTF-8 encoding, with an optional prefix on each line that depends on the style.
For some of the formats, each frontmatter line is prefixed with a prefix to make sure the entire file remains valid in a given syntax (Python, Rust, SQL, etc.). This prefix is stripped during parsing.
It is recommended to use a prefix with a trailing space (such as # or // ) but a bare
prefix without the trailing space (# or ##) is also allowed.
Other whitespace is preserved (before parsing with YAML).
Note that YAML comments, which are lines beginning with # in the metadata, are allowed.
For example, for hash style, this means there must be two hashes (# # or ##) at the
start of a comment line.
There is no restriction on the content of the file after the frontmatter. It may even contain other content in frontmatter format, but this will not be parsed as frontmatter. Typically, it is text, but it could be binary as well.
Frontmatter is optional. This means almost any text file can be read as frontmatter format.
This is a simple Python reference implementation. It auto-detects all the frontmatter styles above. It supports reading small files easily into memory, but also allows extracting or changing frontmatter without reading an entire file.
Both raw (string) parsed YAML frontmatter (using ruamel.yaml) are supported. For readability, there is also support for preferred sorting of YAML keys.
# Use pip
pip install frontmatter-format
# Or poetry
poetry add frontmatter-format
Basic use:
from frontmatter_format import fmf_read, fmf_read_raw, fmf_write, FmStyle, custom_key_sort
# Write some content:
content = "Hello, World!\n"
metadata = {"author": "Test Author", "title": "Test Title"}
fmf_write("example.md", content, metadata, style=FmStyle.md)
# Read it back. Style is auto-detected:
content, metadata = fmf_read("example.md")
print(content) # Hello, World!
print(metadata) # {'author': 'Test Author', 'title': 'Test Title'}The file then contains:
---
author: Test Author
title: Test Title
---
Hello, World!By default, writes are atomic. Key sort is preserved, but you can provide a sorting function if you prefer metadata keys to be in a specific order (often more readable than being alphabetical). There is also an option for making custom sorts, so certain keys come first and the rest are sorted in natural order.
Examples with more formats:
# Write in any other desired style:
html_content = "<p>Hello, World!</p>"
title_first_sort = custom_key_sort(["title", "author"])
fmf_write("example.html", content, metadata, style=FmStyle.html, key_sort=title_first_sort)The file then contains:
<!---
title: Test Title
author: Test Author
--->
Hello, World!YAML parsing is optional:
# Read metadata without parsing:
content, raw_metadata = fmf_read_raw("example.md")
print(repr(raw_metadata)) # 'author: Test Author\ntitle: Test Title\n'
The above is easiest for small files, but you can also operate more efficiently directly on files, without reading the file contents into memory.
from frontmatter_format import fmf_strip_frontmatter, fmf_insert_frontmatter, fmf_read_frontmatter, fmf_read_frontmatter_raw
# Strip and discard the metadata from a file:
fmf_strip_frontmatter("example.md")
# Insert the metadata at the top of an existing file:
new_metadata = {"title": "New Title", "author": "New Author"}
fmf_insert_frontmatter("example.md", new_metadata, fm_style=FmStyle.yaml)
# Read the raw frontmatter metadata and get the offset for the rest of the content:
metadata, offset = fmf_read_frontmatter("example.md")
print(metadata) # {'title': 'Test Title', 'author': 'Test Author'}
print(offset) # The byte offset where the content starts
raw_metadata, offset = fmf_read_frontmatter_raw("example.md")
print(raw_metadata) # 'title: Test Title\nauthor: Test Author\n'-
Hasn't this been done before? Possibly, but as far as I can tell, not in a systematic way for multiple file formats. I needed this myself, and think we'd all be better off if more tools used YAML metadata consistently, so I've released the format and implementation here.
-
Is this mature? This is the first draft of this format. But I've been using this on my own projects for a couple months. The flexibity of just having metadata on all your text files has been great for workflows, pipelines, etc.
-
When should we use it? All the time if you can! It's especially important for command-line tools, AI agents, LLM workflows, since you often want to store extra metadata is a consistent way on text inputs of various formats like Markdown, HTML, CSS, and Python.
-
Does this specify the format of the YAML itself? No. This is simply a format for attaching metadata. What metadata you attach is up to your use case. Standardizing headings like title, author, description, let alone other more application-specific information is beyond the scope of this frontmatter format.
-
Can this work with Pydantic? Yes, definitely. In fact, I think it's probably a good practice to define self-identifiable Pydantic (or Zod) schemas for all your metadata, and then just serialize and deserialize them to frontmatter everywhere.
-
Isn't this the same as what some CMSs use, Markdown files and YAML at the top? Yes! But this generalizes that format, and removes the direct tie-in to Markdown or any CMS. This can work with any tool. For HTML and code, it works basically with no changes at all since the frontmatter is considered a comment.
-
Can this work with binary files? No reason why not, if it makes sense for you! You can use
fmf_insert_frontmatter()to add metadata of any style to any file. Whether this works for your application depends on the file format. -
Does this work for CSV files? Sort of. Some tools do properly honor hash style comments when parsing CSV files. A few do not. Our recommendation is go ahead and use it, and find ways to strip the metadata at the last minute if you really can't get a tool to work with the metadata.
-
Does this also work for YAML files? Yes! It's fine to have YAML metadata on YAML metadata. There are just two nuances.
Firstly, watch out for duplicate
---separators, if you insert frontmatter in front of a file that already has it.Secondly, it's up to you to use the YAML itself to distinguish whether a file has frontmatter or is just a plain YAML file. Both of these can be avoided if you use plain YAML with
---separators only when using frontmatter format.