Senior Vice President of Oracle Database Development Kris Rice published this block of Skills for Oracle AI Database earlier this week. In only a few days, it’s gotten quite a bit of interest!
If you’re new to Skills in general, I’ve published two posts on the subject, with examples of how to take advantage of it with our MCP Server.
- Creating a Skill for Loading CSV to Oracle via SQLcl
- Introducing Kris’ skills with 2 examples – find me the worst SQL in my db for the past day, and help me find security issues in my database
But that’s just skimming the surface.
“Just skimming the surface”
I think that describes where many of us are in our AI journey. You’re sitting at a prompt of some sort, and you ask a question. And after some back and forth, you get an answer. It’s even, a pretty good answer!
But.
Are you thinking strategically? Let’s look at one of the exercises I went through yesterday, taking a look at possible security vulnerabilities in my database.
This is actually a pretty bad situation. And I know about it, because I asked. But, if I want to know about it tomorrow, and the day after, I have to ask, again.
Using an Agent like it was meant to be used
Let’s get out of ‘Ask’ mode, and let’s get our money’s worth from our AI ‘buddy.’ To start things off, I asked what were some good tasks I could tackle now, as a DevOps Engineer or DBA, that would really take advantage of these newly imported Oracle skills.
Because let’s face it, even though I’ve been doing this for decades, sometimes I can’t get out of my own way, my own biases, and my own habits. LLMs aren’t encumbered by this so much. I mean, they can be, our prompts will influence them, but let’s assume you are asking straightforward questions.
now that we have our new skills for oracle and we have an mcp server for oracle available to us…
What are the top 10 to 15 types of tasks an oracle database administrator or devops engineer could accomplish with their Agents (like Claude!), faster, more efficiently, and easier?
Here’s what it came up with –
And I quickly zeroed in on this bit –
Now the real work begins
great, now take 1 of these, and turn it into new task where a dba could have the process automated, for example
create a simple web app.
data will be retrieved by a set of ORDS REST APIs, and the dba will have the simple web app running local on their machine
Let’s create the database bits, the ords rest apis, and the ‘localhost’ web app the dba can open to get a list of available reports and those details
Here’s what I saw this morning, after letting it run.
Now I can just browse the latest findings, on demand. My agent created for me:
- a set of tables to store our data
- a pl/sql api to do the gathering and analysis
- a scheduler job to gather information from the database’s Audit trail
- a set of ORDS REST APIs to deliver this information to our web app
- the actual web application i can startup with a single command
And here’s what I needed to do to make it work:
- install those skills for my agent – more on that in like 2 seconds!
- have an MCP Server (SQLcl) so it could get to the database and install these objects
- do about 5 minutes of back and forth helping it fix the PL/SQL package
- granting access to the dictionary views required to get this information for the application schema user
- that’s it, literally that’s it – I spend maybe 20 minutes on this task
As a test, I created a ‘bad user’ with a really, really bad password, and guessed my way into the database. This showed up immediately on the report.
Getting started is super easy: Installing those skills
A very cool resource is skills.sh
Assuming you have node.JS on your machine, you can simply run this package manager command to get going –
npx skills add krisrice/oracle-db-skillsMeanwhile, back here, down on Earth
Yes, I live in the real world, and I know that AI is not a Silver Bullet. But it’s an amazing helper, if not even a ‘doer.’ However, some words of caution to accompany my super encouraging post:
- we can’t trust these agents, I did have to intervene several times to provide guidance/corrections
- the AI Skills will cut down on how often you need to do this, but it won’t take away your personal responsibility – you still need to be an active participant
- security when it comes to allowing your Agent to interact with your database via your MCP Server remains paramount
- ‘Jeff, your magic AI is just running scripts. I have had these scripts for 15 years, this gets me nothing.’ To that. I would say, the Agents can run your scripts faster, and ALSO analyze the results, and action responses to the findings. Your scripts are AWESOME, so level up and import them as Skills so your Agents can use your scripts as well.
- To previous point, see this quick read from our own Miguel Sánchez Villafán , “AI skills vs. scripts: when to use each”
Behind the scenes
While note the point of this post, I’m not here to show or share the code vs inspiring you to do more, I do realize you might be curious. Here’s a peek at what my Agent put together, via our Skills and MCP Server.
Not shown, the DBMS_SCHEDULER job, which simply runs daily each morning at 7AM, and calls the PL/SQL api to get the latest findings.
