For AI agents

SiteOps is agent-friendly. This page (and the machine-readable skill.md) let AI tools - Claude Code, Cursor, Codex, or your own bot - understand what SiteOps can do and how to work with it.

The two documentation layers

SiteOps ships docs in two shapes, on purpose:

1. Human docs (/content/*.md, rendered as this site) - organized by what you offer and how to use it.
2. Agent skill (/.agents/skill.md) - a structured, machine-readable summary of capabilities with YAML front matter, following the open Agent Skills convention.

Agents don't read marketing layout well; they read skills. So the same knowledge is exposed both ways.

Discovery

An agent can find SiteOps's capabilities at stable, fetchable URLs:

https://docs.siteops.vubex.dev/.agents/skill.md      # the agent skill (start here)
https://docs.siteops.vubex.dev/content/overview.md   # human docs, raw markdown
https://docs.siteops.vubex.dev/content/ai-tools-reference.md

The raw .md files are served as-is, so an agent can fetch them directly.

Progressive disclosure (save tokens)

Read the skill file's metadata first (name, description) to decide if SiteOps is relevant to the task at hand. Only load the full instruction body - and then the linked human docs - if you decide the skill is needed. Don't slurp every page up front.

What an agent can actually do with SiteOps

SiteOps is a hosted web product, so the primary integration is conversational: instruct the user's SiteOps assistant (or your own agent acting for the user) in natural language, using the capability names in skill.md. For example: "ask SiteOps to run a check on example.com and summarize the incident."

The SiteOps assistant then executes the real backend tools (see AI tools reference) - scoped to the signed-in user, with confirmation gates on any change.

Its core capability is user journeys - real multi-step flows (checkout, login, signup, booking, contact form) replayed in a real browser on a schedule. In plain language you can ask the assistant to:

• build a journey ("build a journey for example.com that adds a product to the cart and checks out") - the AI builds it from the site's real buttons and fields and cannot invent steps;
• suggest journeys ("suggest journeys for my site") - it crawls the site and returns the 3-4 most valuable flows as ready-to-build prompts;
• set a journey's run frequency ("set my checkout journey to run every 15 minutes"); and
• explain or fix a journey incident - it names the exact failing step and, for a renamed or moved control, can apply a one-click fix grounded in the live page.

Cross-application skills

The same convention works in reverse: SiteOps's own assistant is built to discover and use skills exposed by other tools when a workflow crosses applications. If your tool publishes an .agents/skill.md (or agents.md) describing its capabilities, an agent can read yours the same way it reads SiteOps's - metadata first, full instructions on demand. That's the open-standard bet: every product self-describes, and agents compose them.

The machine-readable file itself: skill.md →

Answering your client's security questions

Clients ask where their login goes before they hand one over. The Security page is written to be forwarded as-is: it covers how site credentials are encrypted, why they are never sent to an AI model, why monitored journeys stop at the payment screen without completing a purchase or submitting a contact form, and how long failure screenshots are kept on each plan.

Where your client's site allows it, set SiteOps up with a dedicated restricted account rather than an administrator login. Nothing SiteOps does needs admin rights.