Agent Skills

sitemd ships twelve skills — structured task definitions that encode multi-step workflows with tool orchestration, voice rules, and quality gates. Type /skillname in any supported agent to run them. Skills go beyond context files by encoding complete procedures, not just reference information.

Quick reference

Skill Usage What it does
/write /write [type] [topic] Generate content (page, docs, blog, changelog, roadmap)
/launch /launch [demo|scratch] Start the dev server with live build sync
/deploy /deploy Build and deploy to configured hosting target
/kickstart /kickstart [description] Generate a complete first-draft site from one prompt
/clone /clone [url] Scrape and import an existing website
/feedback /feedback [type] [title] Submit a bug report, feature request, or question
/seo /seo [scope] Run SEO health check with scored report and fix suggestions
/og-image /og-image Generate a branded OG social share image from theme colors
/import /import [file path] Import a raw markdown file as a sitemd page
/site /site Show project status and available actions
/reload /reload Trigger a dev server hot-reload without restarting
/shutdown /shutdown Kill the dev server (user-invoked only)

/write

Generate high-quality site content. The first argument is the content type — page, docs, blog, changelog, or roadmap — and everything after is the topic. Each type has its own structure template, voice rules, and quality gates.

The skill gathers site context via MCP tools before writing, matches the tone of existing content, validates the result, and handles group membership and sidebar ordering automatically. When invoked mid-conversation with no arguments, it assumes you want docs or a changelog covering the work just completed.

Content is structured for both human readers and AI citation. The skill uses question-based headings ("How does X work?" rather than "X Mechanics"), starts every section with a standalone answer sentence that works as an extracted citation, includes updated: dates as freshness signals on docs, and favors verifiable data over vague claims in blog posts.

See Content Generation for the full reference — content types, MCP tools, draft mode, and automation settings.

/launch

Start the dev server with file watching and live build sync.

/launch              # serve your project
/launch demo         # serve the demo component showcase
/launch scratch      # serve the blank-slate template
Target Directory Default port
(none) sitemd/ from settings/build.md (default 4747)
demo sitemd/templates/demo/ from its settings/build.md (default 4848)
scratch sitemd/templates/scratch/ from its settings/build.md (default 4747)

The skill reads the port from the target's settings/build.md, kills any existing process on that port, starts the server in the background, and opens the browser automatically.

File changes auto-rebuild — edit a page and the browser refreshes. The companion skill /shutdown kills the server when you're done.

/deploy

Build the site and publish it to the configured hosting target in one step. The skill runs pre-flight checks before deploying:

  1. Authentication — verifies you're logged in. If not, initiates the browser login flow.
  2. Domain — checks that a domain is configured in settings/deploy.md. Prompts you to set one if missing.
  3. Credentials — ensures deploy target credentials are stored in the config store.
  4. Deploy — calls sitemd_deploy to build and publish.

Four hosting targets are supported, each with its own required credentials:

Target Required config keys
Cloudflare (default) deploy.cloudflareProject, deploy.cloudflareAccountId, deploy.cloudflareApiToken
Netlify deploy.netlifySiteId, deploy.netlifyToken
Vercel deploy.vercelProjectId, deploy.vercelToken, deploy.vercelTeamId (optional)
GitHub Pages deploy.githubRepo, deploy.githubToken, deploy.githubBranch (optional)

Credentials are stored via sitemd config set or the sitemd_config_set MCP tool — never in settings files. Set the target in settings/deploy.md frontmatter.

See Deploy for hosting setup details and CI configuration.

/kickstart

Generate a complete first-draft site from a single prompt. This is the fastest way to go from zero to a working site.

/kickstart a developer tools documentation site
/kickstart landing page for a coffee subscription service

If no argument is provided, the skill asks: "What are you building? One sentence is enough."

The skill plans 3–6 pages (always including a home page), then executes immediately without waiting for confirmation — this is a first draft, not a contract. It:

  1. Updates settings/meta.md, settings/header.md, settings/footer.md, and settings/groups.md to match
  2. Creates each page with real content and rich components (buttons, cards, embeds)
  3. Validates every page after creation

With the dev server running (/launch), pages appear in the browser as they're created. The default welcome page is removed before the new home page is written.

/clone

Scrape an existing website and import its content into a sitemd project. The clone tool extracts text, images, navigation, forms, embeds, and card layouts — mapping each to sitemd components. Styling is not copied; sitemd's themes apply.

See Website Cloning for the full reference — page discovery, content extraction, asset handling, and the clone report format.

/seo

Run a comprehensive SEO health check across site settings and all pages. Returns a scored report (0–100) with errors, warnings, and info-level opportunities. Each finding includes an educational explanation of why it matters and a specific fix suggestion.

Pass an optional scope: /seo settings (site-level only), /seo pages (page-level only), or /seo /about (single page). The agent presents the full report and offers to fix issues — updating settings files, adding missing frontmatter fields, or rewriting duplicate descriptions.

The audit also runs automatically during /deploy, presenting the report before proceeding and letting you fix issues or skip.

See SEO Audit for the full check catalog, scoring formula, and fix workflow.

/og-image

Generate a branded Open Graph social share image using your site's theme colors and content. The skill reads your config, asks three questions (heading, subheading, layout), generates a 2400×1260 PNG, and saves it to media/og-image.png.

Two layouts: simple (centered text with accent bar and brand logo) and mockup (browser chrome with split-pane markdown editor and rendered preview). Colors are resolved from your theme/styles.css for the chosen theme mode — dark, light, or paper.

The skill automatically updates settings/seo.md to use the custom image and disables auto-generated OG cards. Iterate by asking for changes — the agent regenerates without needing a server restart.

See Custom OG Images for the full reference — color resolution, MCP tool parameters, and settings behavior.

/feedback

Report bugs, request features, or ask questions from any AI agent. The skill collects environment context (sitemd version, Node version, OS) and opens a pre-filled GitHub issue in your browser. If an error occurred earlier in the conversation, it auto-populates the body.

See Feedback for the full reference — issue types, CLI command, and MCP tool parameters.

/import

Import a raw markdown file into the site as a properly formatted page.

/import notes/api-guide.md
/import ~/Downloads/blog-post.md

The skill reads the source file, generates frontmatter (title with site suffix, description under 160 characters, slug), and determines group membership based on content topic. During conversion:

The skill never overwrites existing files — if the target path already exists, it reports the conflict and asks what to do. Content is preserved faithfully; the goal is format conversion, not rewriting.

/site

Show project status and available actions — the MCP-powered equivalent of the interactive sitemd CLI.

Fresh projects (site name "My Site" and 4 or fewer pages) trigger the onboarding flow:

  1. Launches the dev server with /launch
  2. Launches the demo showcase with /launch demo
  3. Asks what you're building
  4. Runs /kickstart with your answer

Existing projects display a formatted status overview:

The skill lists available actions as numbered options and asks what you'd like to do next.

MCP tools used by /site

Action MCP tool
Project status sitemd_status
Deploy sitemd_deploy
Log in sitemd_auth_login
Poll login sitemd_auth_poll
Auth status sitemd_auth_status
Log out sitemd_auth_logout
View config sitemd_status
Set config sitemd_config_set
Delete config Edit .sitemd/config.json

/shutdown

Kill the dev server. This is the companion to /launch.

The skill discovers all ports by scanning settings/build.md files across the project, then kills any process running on each port. If no build settings are found, it falls back to port 4747.

User-invoked only. The /shutdown skill has a special flag (disable-model-invocation: true) that prevents agents from calling it automatically. It only runs when you explicitly type /shutdown. This prevents agents from accidentally killing your dev server during other workflows.

Cross-agent compatibility

Skills ship in two directories with identical content:

Directory Discovery
.claude/skills/ Claude Code
.cursor/skills/ Cursor
.agents/skills/ Codex CLI, SkillsMP, any SKILL.md scanner

GitHub Copilot discovers skills via .github/skills/ (populated from the same source). Cursor also gets three auto-attaching rules in .cursor/rules/ for context that loads based on which files you're editing. All skill directories contain the same SKILL.md files — the skill procedures, MCP tool calls, and voice rules are identical regardless of which agent runs them.

Skill file format

Each skill is a single SKILL.md file with YAML frontmatter and a markdown procedure body:

---
name: skillname
description: One-line description shown in skill discovery.
argument-hint: "[type] [topic]"
---

Frontmatter fields:

Field Required Description
name yes Skill identifier — maps to /name invocation
description yes Shown in agent skill listings and help text
argument-hint no Placeholder text showing expected arguments
user-invocable no When true, skill appears in user-facing command lists
disable-model-invocation no When true, agents cannot call this skill automatically

Body structure: