Skip to main content

Documentation Agent Instructions

Guidelines for AI agents editing files in docs/.

Conventions

  • Pure .md files, no .mdx or Mintlify components
  • Frontmatter: only title (required by Mintlify)
  • All navigation structure lives in docs.json
  • Archive of old docs in docs/archive/ for reference

Writing Standards

Content

  • Write for future-me — these are personal reference docs, not tutorials
  • Optimize for evergreen content — things that change frequently belong in the blog, not reference pages
  • Wrong docs are worse than no docs — if something’s outdated and can’t be fixed soon, remove it
  • When a section accumulates too many caveats, rewrite it from scratch

Style

  • Be concise — get to the point
  • Active voice: “Traefik routes traffic to…” not “Traffic is routed by Traefik…”
  • Consistent terminology — pick one name per thing and stick with it
  • Specify language on code blocks (yaml, bash, etc.)
  • Use mermaid for diagrams with complex relationships or branching logic
  • Use nested bullet lists for simple hierarchies and linear flows
  • No ASCII art diagrams

Structure

  • One H1 per page (the title frontmatter handles this — don’t add another # heading)
  • Sequential heading hierarchy — don’t skip levels (e.g. ## to ####)
  • Keep content scannable: short paragraphs, bullet points, tables for structured data
  • Unique heading names within a page

Linking

  • Root-relative paths for internal links: [hardware](/infrastructure/hardware)
  • Descriptive link text — “see backups” not “click here
  • Anchor links auto-generate from headers: ## External Drives#external-drives

Media

  • Only add screenshots/diagrams when text alone isn’t clear enough
  • Alt text on all images
  • Media is supplementary — the text should stand on its own

AGENTS.md Per-Service Files

  • Keep files concise — if it’s intuitive for an agent to figure out, it shouldn’t be documented
  • Never hardcode static info that can be discovered at runtime (env var names, ports, IPs, URLs)
  • Agents should read compose files and .env files to discover those values
  • No file trees, no how-tos, no method documentation — point to --help instead
  • Only document: interface type (CLI/API/config-file), auth pattern, and non-obvious quirks