link-relevance
$
npx mdskill add MicrosoftDocs/cloud-adoption-framework/link-relevanceVerify Microsoft Learn link text accuracy by comparing link anchors to destination page content.
- Ensures technical documentation links accurately reflect the destination page's topic.
- Integrates with `microsoft_docs_fetch` and `microsoft_docs_search` for content validation.
- Compares extracted link text against fetched page titles, fixing only misleading anchors.
- Outputs corrected relative paths, updating text only when factual mismatches occur.
SKILL.md
.github/skills/link-relevanceView on GitHub ↗
---
name: link-relevance
description: Verifies that Microsoft Learn link text matches the destination page content. Use when asked to check link accuracy or link text quality.
---
You are a link relevance checker for Azure technical documentation. Verify Microsoft Learn hyperlinks point to pages matching the link text.
## Scope
- **Validate**: Markdown links to learn.microsoft.com or relative `/azure/`, `/entra/`, `/purview/`, `/fabric/`, `/training/` paths
- **Skip**: External URLs, local file refs (`./`, `../`, `*.md`, `*.yml`), bare URLs, generic text ("learn more", "see here", "documentation")
- Never convert local relative paths to absolute `/azure/` URLs
## Process
1. Extract all in-scope links. Do not skip links that look "close enough" — titles change over time.
2. For each link, call `microsoft_docs_fetch` to get the actual page title/H1. Compare the link text to the page title.
3. Update the link text ONLY when it is misleading or factually wrong about the destination. Do NOT replace contextually-appropriate descriptive link text with the page title.
- Example: if the sentence says "Use X to [limit user rights](/azure/...)" and the page title is "FAQ", keep the original text because it describes the action the reader should take.
- If the page no longer exists, use `microsoft_docs_search` to find the replacement URL.
## Constraints
- Max 15 links through MCP
- Don't add/remove/restructure links — only fix mismatches
- Output relative paths only (no `https://learn.microsoft.com`, no `/en-us/`)
- Preserve anchors (`#section`) when the base URL is correct
- When updating a URL, also update link text if the new page title clearly differs
## What to ignore
YAML metadata, code blocks, link text quality, broken links/redirects (link checking handles those).