SuperAPI AI Directory

frontmatter-description

Writing & research

View repository

Updated 10-16 00:32

About this skill

Frontmatter Description

Check and optimize MetaDescription frontmatter fields in markdown documentation files. Produces actionable fixes for descriptions that violate the rules below.

When to Use

  • Adding or editing a documentation page and need to write a MetaDescription.
  • Auditing existing pages for SEO or discoverability improvements.
  • Asked to check, review, or optimize frontmatter descriptions.
  • Editing content articles and want to ensure the MetaDescription reflects the updated content.

Rules

Every MetaDescription value must satisfy all of the following:

Rule Detail
Length Maximum 160 characters.
Tone Action-oriented, value-focused, factual, and impersonal.
Voice No "you can", "users can", or "this page explains".
Sentences Complete sentences, not fragments or labels.
No colons Do not include : in the value — it breaks YAML parsing.
Uniqueness Each description must be unique across the docs.
Reusable variables Preserve existing {% data variables.<group>.<name> %} references. Do not replace them with rendered product names.

Descriptions must also reflect the page's primary persona, reader intent, and article purpose. Establish that framing by following the content-framing guidance before suggesting or applying a description. If the existing page and writer's request support multiple materially different interpretations, ask the writer to confirm the framing before editing.

Procedure

1. Identify target files

Determine which files to check:

  • If given specific files, use those.
  • If asked to audit a folder (e.g., docs/copilot/), find all .md files in that folder recursively.
  • Skip files that have no YAML frontmatter (e.g., README.md files without --- delimiters).

2. Extract and validate

For each file, read enough of the page to identify its primary persona, reader intent, and article purpose. Then read the YAML frontmatter and check the MetaDescription field:

  1. Missing — Flag if MetaDescription is absent.
  2. Length — Flag if the value exceeds 160 characters. Report the current length.
  3. Voice — Flag if it contains "you can", "users can", "this page explains", or similar reader-addressing phrases.
  4. Tone — Flag if the description is a fragment, a label, or passive rather than action-oriented.
  5. Context — Flag if the description does not mention the relevant feature or tool.
  6. Content alignment — Flag if the description does not represent the page's purpose or the outcome sought by its primary persona.
  7. Forbidden words — Flag any occurrence of "teaching", "enable", "disable", or condescending terms.
  8. Colons — Flag any : character in the value.
  9. Versions — Flag any version numbers (e.g., "v1.90", "VS Code 1.90").
  10. Formatting — Flag Jinja2 variables ({{...}}), HTML tags, or Markdown formatting. Do not flag supported reusable data variables ({% data variables.<group>.<name> %}).
  11. Reusable variables — When a description contains a reusable data variable, verify that its full path exists in data/variables and resolves to the intended text. Preserve valid variables exactly as written.

3. Report findings

Present results as a table:

| File | Issue | Current value | Suggested fix |
|------|-------|---------------|---------------|
  • Group by file path.
  • For each issue, provide the current MetaDescription and a rewritten suggestion that passes all rules.
  • If a description passes all checks, omit it from the table (or mark it as passing if the user asked for a full report).

4. Apply fixes (when asked)

  • Edit the MetaDescription value in the YAML frontmatter.
  • Do not change any other frontmatter fields.
  • Do not alter the page body content.
  • Preserve existing reusable data variables. Never replace a variable with its rendered product name.
  • Verify the fix passes all rules and reflects the confirmed page framing before applying.

Examples

Good descriptions:

  • Get AI-powered inline suggestions from GitHub Copilot in VS Code, including ghost text completions and next edit suggestions.
  • Configure and manage extensions in Visual Studio Code to customize your development environment.
  • Debug Python applications in Visual Studio Code with breakpoints, variable inspection, and integrated terminal output.
  • Build a web app with an AI agent in {% data variables.product.prodname_vscode_shortname %}, then review and validate the result.

Bad descriptions and why:

Description Problem
This page explains how to use Copilot. Uses "this page explains".
You can debug your code with VS Code. Uses "you can".
Copilot Fragment, not a sentence.
Learn how to enable the new feature in VS Code 1.90. Uses "enable" and includes a version number.
Set up debugging: configure launch.json for Python. Contains a colon.