Diagnose a site
Have an AI assistant find the likely cause of a problem on a site, with evidence, without changing anything.
Before you start
- An AI assistant is connected to your organisation: see Connect an AI assistant.
- A description of the symptom, up to 2,000 characters: what you see, on which address, since when.
Start the diagnosis
Describe the problem as you would to a colleague. The assistant follows the diagnose-site playbook, which only reads.
Use the Corefinity MCP prompt "diagnose-site" with site: shop.example.com,
symptom: the checkout has answered 502 since about 10:40 this morning.
Follow it step by step and tell me before anything that changes my account.Add environment: Staging to look at an environment other than Production. When a check needs a scope the token lacks, you are asked to allow it: see Scopes.
What the assistant reads
| Check | What it reads | See it yourself |
|---|---|---|
| State | The site, its live state and any changes waiting for the maintenance window. | The site's page |
| What changed | Operations, commands, deployments and the assistant's own work, in one feed. | Activity |
| Logs | When errors rose, then the web server, PHP and application logs around that time. | Logs |
| Requests | The address you reported, the site's key pages, and the error rate and response times over the period. | Traffic |
| Resources | Memory and disk over the last 24 hours. | Resources |
Only when these do not explain the problem does it run commands on the site, and only ones that change nothing, such as wp core verify-checksums or df -h. For a site in a CoreCluster it also reads the other sites' logs, when one of them may be the cause.
What you get
- The likely cause, with the log lines, timings and usage figures behind it.
- The least risky fix. You decide whether to make it: the assistant changes nothing in this job.
- What to use next: Recover a site to undo a bad change, or Create an environment to try a fix away from production.
Related
Was this helpful?