You've seen how to ask the MCP questions, pull a report, even build a dashboard. Skills are how you stop doing all that from scratch each time. A Skill packages a whole job (the questions to ask, the order to read the data, the deliverable to produce) so you can run it in a sentence instead of a long prompt.
Think of it like this: the MCP provides the data, Skills provide the process. The MCP reads your Sitebulb audits; a Skill is the method wrapped around that data that turns it into a decision or a deliverable.
Not connected yet? Everything on this page assumes the Sitebulb MCP is already hooked up to Claude or ChatGPT. If it isn't, Sitebulb MCP: Start Here gets you connected in a couple of minutes, then Getting Started with the Sitebulb MCP covers your first prompts.
What a Skill is
A Skill is a reusable set of instructions that tells the assistant how to do a specific job. You capture how the work should be done once, and from then on the assistant follows it.
This is what makes Skills deterministic. Prompt from scratch and you get a slightly different answer every time, because you word it differently and the assistant improvises differently. A Skill takes that variability out: the same questions, the same order, the same shape of output, to the same standard on every run. "Prioritise the issues in this audit" stops being a fresh gamble each time and instead becomes a repeatable process.
That's also where the value sits. A Skill lets you embed your process and your preferences: your definition of a quick win, the way you structure a report, the tone you take with a client, the thresholds you actually watch. Bake it in once, and every output carries it without you re-explaining.
Once you've got your head around how they work, we strongly urge you to build out your own Skills, but to get you going, we've built some you can use out of the gate.
Three things run through all five we've built so far:
They interview before they answer. Business context leads the decision, so the same crawl produces a different answer for an ecommerce site mid-migration than for a publisher in a traffic slump.
They reframe, never fake. The scores stay honest. Context only decides what leads, and the Skill tells you when it's demoting something ("this affects every page, but it's best practice with no ranking effect").
They hand off, never dead-end. Every output assumes a next step, whether that's "re-brand this in our styling" or "run this weekly and post it to Slack".
You do a little up-front work to get a Skill set up, then it pays back every time. Over time, the real power is writing your own, for the jobs you do and the way you do them.
The five Skills we've built
What To Fix First - triage, for everyone's first five minutes with a crawl.
What Changed - for agency account managers on a retainer.
Technical SEO Audit - a client deliverable, for consultants and agencies.
Release Check - for in-house teams shipping website changes on a release cadence, to see if anything slipped.
Dev Handoff - for pushing issues into Jira to get the fixes shipped.
There's a fuller description of each below.
How to get them
(GitHub link to add once the repo is live.) All five live in the Sitebulb Skills GitHub repository. Each Skill is a folder containing a SKILL.md file (the instructions the assistant reads) plus any supporting files. Download the one you want, or grab the whole set.
How to install them
No terminal needed, and the same Skills work in both Claude and ChatGPT.
Claude
Head to Customize, navigate to 'Skills' and then Add -> Upload a skill
Then just upload the folder you downloaded. Claude reads the SKILL.md, shows you a summary of what the Skill does, and you switch it on.
For Claude Code (if you prefer the terminal) - drop the Skill into your skills folder and call it with a slash command.
ChatGPT
Head to Plugins, navigate to the 'Skills' tab and then Add -> Upload a skill
Then just upload the folder you downloaded. ChatGPT reads the SKILL.md, shows you a summary of what the Skill does, and you switch it on.
Once a Skill is installed, you don't need a long prompt. Trigger it by name ("run What Changed on the Acme Store project"), call it by a slash command (in Claude only) or just describe what you want and let the assistant pick the right one.
What each Skill does
What To Fix First
The one for your first five minutes with any crawl. It asks a couple of questions about the site and your situation, then ranks the issues by impact against effort and hands back the three to five genuinely worth doing first. Think of it as a second opinion from a senior consultant, on tap. It's interactive, so you walk through it.
Example: "Run What To Fix First on the 'Initial Client Audit' project and lay it out as an impact-vs-effort quadrant."
What Changed
Built for the monthly retainer conversation. It compares the two most recent crawls and tells you what changed: which things got better, which things got worse, then a watch-list, without you even opening Sitebulb. Handy whether you're writing the report or already on the way to the call...
Example: "Run What Changed on the 'Acme Store' project, comparing this month's crawl to last month's, and show me the wins, regressions and watch-list."
Technical SEO Audit
Turns a crawl into a client-facing audit document: the five to ten issues that actually matter for that engagement, written for the person who'll read it. It arrives as a working draft with review comments pinned where your judgement is needed, so you're reviewing and signing off, not starting from a blank page. And by the way it really is important that you DO review and amend the doc, since it's your name that's going on it!
Example: "Run Technical SEO Audit on the 'Acme Ltd' project".
Release Check
Designed to be the smoke alarm for teams shipping website updates regularly. A noindex ships by accident, canonicals vanish from a template, a page type loses its H1, and you find out weeks later when traffic slides. Set the watch-list and thresholds once, tell it what you changed on purpose so deliberate changes don't cry wolf, and most Tuesdays it just Slacks you a two-line "all clear". Pay attention when it doesn't.
Example: "Run Release Check on 'Weekly Sitebulb'. Yesterday's release swapped the product-page template, so flag anything off the watch-list and post the result to Slack."
Dev Handoff
Closes the gap between finding an issue and getting it fixed. It writes proper tickets straight into Jira, Linear, Asana, Monday or ClickUp, each with what's wrong, why it matters, sample URLs and an acceptance criterion that doubles as its own test. Every ticket is previewed in the chat before anything is created.
Example: "Run Dev Handoff on 'Initial Client Audit' and raise the canonical-to-noindex issue as a Jira ticket for the dev team."
Automating your Skills
Some of these Skills naturally make sense to automate, but others not so much. Two are interactive by design (What To Fix First and Technical SEO Audit): their value is the conversation and your judgement, so you're meant to be in the room. Two run the moment a scheduled crawl lands (What Changed and Release Check). Dev Handoff sits in between, on demand behind a preview gate.
The one constraint: the MCP is read-only and can't start a crawl. So automation always means the crawl is already scheduled in Sitebulb, and something that can run on a schedule (Claude Cowork) runs the Skill when it finishes. Two that work well:
The monthly report drafts itself. The overnight crawl finishes, Cowork runs What Changed, and posts the wins to the client's Slack. You open your laptop to a draft, not a blank page.
The Tuesday all-clear. You ship Monday, the crawl runs that night, and Cowork runs Release Check against your watch-list on Tuesday. If nothing broke, you hear nothing.
Build your own
The five are a starting point, not the ceiling. The moment you catch yourself typing the same long prompt for the third time, that's a Skill waiting to be written. Describe the job, the steps you follow and the output you want, and Claude will help you draft the SKILL.md for you to refine.