Sidematter format is a simple, universal convention for keeping metadata and assets alongside a primary document. It is a useful complement to frontmatter format.

Many tools and formats need structured data associated with a document but not inside it:

• Metadata that can’t easily be stored in a document itself, due to size or because it is updated more often (e.g. document annotations, full version history, etc.)

• Additional files that must travel with a document, such as images or resources associated with an HTML or Markdown file.

Sidematter format defines a minimal set of conventions for naming and resolving such “sidecar files” in a consistent way.

Sidecar patterns are often used in data pipelines, in exports from web browsers, and other applications. Unfortunately, there’s no consistent convention for naming and organizing such external files, leading to varied, ad-hoc approaches that don’t interoperate well.

This repository is a description of the format and a reference implementation. The implementation is in Python but the format is simple and can be adopted by any tool or language.

Sidematter format does not specify a way to bundle the outputs, but a file plus its sidematter files can easily be bundled together in a zip or tarball.

Sidematter format is easiest to illustrate by an example. Given a primary document report.md, some possible sidematter files would be:

report.md
report.meta.yml        # YAML metadata
report.assets/         # Asset directory
    figure1.png
    diagram.svg
    styles.css

The document can reference assets with relative paths:

# My Report

![Key findings](report.assets/figure1.png)

See the [full diagram](report.assets/diagram.svg) for details.

More examples across different file types:

blog-post.md           → blog-post.meta.yml, blog-post.assets/
index.html             → index.meta.json, index.assets/
presentation.slides    → presentation.meta.yml, presentation.assets/
notebook.ipynb         → notebook.meta.yml, notebook.assets/

Example metadata content:

# report.meta.yml
title: Q3 Financial Analysis
author: Jane Doe
created_at: 2024-01-15
tags:
  - finance
  - quarterly
  - analysis
processing_history:
  - step: data_extraction
    timestamp: 2024-01-15T10:30:00Z
    tool: custom_extractor_v2.1
  - step: analysis
    timestamp: 2024-01-15T11:45:00Z
    tool: pandas_analyzer
source_files:
  - raw_data.csv
  - market_trends.json

If desired, simple YAML metadata could instead be inserted as frontmatter on the file itself, and omitted from the sidematter:

report.md              # Main file with frontmatter format metadata in YAML
report.assets/         # Asset directory with extra files
    figure1.png
    diagram.svg
    styles.css

The implementation there looks for both formats, so will read the metadata on either of these layouts seamlessly.

• Clean separation: Keep metadata and assets separate from primary content, not requiring that it be bundled (like a zip file).

• Predictable asset filenames and metadata syntax: Sidematter files should be auto-detectable via consistent naming convention so it is easy for tools to discover metadata and asset files.

• Schema- and format-neutral: This is a simple convention for attaching metadata and assets. It aims to be flexible and unopinionated. so does not specify any specifics on asset file formats or metadata schema (other than the use of YAML or JSON). The convention works with any file format since sidecars don’t modify the original document.

The sidematter format defines naming conventions for files and directories related to a base document, which can be any file, with any name.

Given a base document with filename basename.extension, the sidematter files are:

• Metadata files: basename.meta.json or basename.meta.yml

• Asset directory: basename.assets/ (directory containing related files)

The sidematter names are formed by dropping the final extension from the base document name, then appending the sidematter suffix:

• Files without extensions get sidematter suffixes directly: README → README.meta.yml, README.assets/.

• For files with multiple extensions (e.g., data.tar.gz), only the final extension is dropped: data.tar.gz → data.tar.meta.yml.

• The schema of metadata files is free-form and tool-dependent. Common metadata conventions include standard fields like title, description, author, created_at, and tags, but applications are free to define their own schemas.

• Both JSON and YAML are allowed. JSON is often preferred for machine-generated metadata due to ubiquitous parsing support. YAML is often better for human-authored or human-readable metadata due to readability and comment support.

• If there is a schema associated with the metadata, follow the standard convention of linking to it with the $schema key, so that tools like IDEs can validate the schema.

In most cases, metadata should only reside in one place, typically basename.meta.yml. Implementations should observe precedence and pick metadata from the first location found in this order:

1. Metadata JSON: basename.meta.json

2. Metadata YAML: basename.meta.yml

3. Optionally, implementations can look for frontmatter on the file itself (if it is a text file)

• Any asset files are allowed: The .assets/ directory structure is free-form and tool-dependent. Files can be organized in subdirectories as needed.

• Relative path resolution: References from the base document (such as Markdown or HTML) to assets should use relative paths starting with the asset directory name.

The Python implementation provides a simple reference implementation for reading and writing sidematter.

from sidematter_format import Sidematter

# Read all sidematter for a document by checking the filesystem.
# Returns an immutable ResolvedSidematter.
paths = Sidematter(Path("report.md")).resolve() 
print(paths.primary)  # Path('report.md')
print(paths.meta)  # {'title': 'Q3 Report', 'author': 'Jane Doe', ...}
print(paths.meta_path)  # Path('report.meta.yml') or None
print(paths.assets_path)  # Path('report.assets') or None

from sidematter_format import Sidematter

# Create a Sidematter object for read/write operations
sm = Sidematter(Path("report.md"))

# Write metadata as YAML (default)
metadata = {
    "title": "Q3 Financial Analysis",
    "author": "Jane Doe"
    "tags": ["finance", "quarterly"]
}
sm.write_meta(metadata)

# Write metadata as JSON
sm.write_meta(metadata, fmt="json")

# Write pre-formatted YAML/JSON string
sm.write_meta("title: My Report\nauthor: Jane Doe\n")

# Remove all metadata files
sm.write_meta(None)

# Get the path for an asset (creates .assets/ directory)
chart_path = sm.asset_path("chart.png")
# Returns: Path('report.assets/chart.png')

# Copy a file into the assets directory
sm.add_asset("~/Downloads/chart.png")
# or with a custom name:
sm.add_asset("~/Downloads/fig1.png", dest_name="chart.png")

# Check if assets directory exists
if sm.resolve_assets():
    print(f"Assets found at: {sm.assets_dir}")

• Hasn’t this been done before?

  Similar patterns exist in various tools (Jekyll’s _files/ directories, Hugo’s page bundles, etc.), but there’s no universal convention that works across different tools and file types. This format provides a simple, consistent approach.

• When should I use sidematter vs frontmatter?

  Use frontmatter format for small, essential metadata, especially on text files of any kind. Use sidematter format for larger metadata, when the original file cannot be modified or metadata that should be updated separately from the original file, or if additional asset files are needed.

• Does this work with version control?

  Yes, of course. Typically you would check in both metadata and assets, but if it’s easy to put *.assets/ or *.meta.json or *.meta.yml in .gitignore to avoid including them in version control.

• Can I use both YAML and JSON metadata?

  Yes, tools should support either or even both formats simultaneously. The convention is that if present, JSON will be used first, since that is often auto-generated or faster to parse.

For how to install uv and Python, see installation.md.

For development workflows, see development.md.

For instructions on publishing to PyPI, see publishing.md.

This project was built from simple-modern-uv.