Most people pose one-off queries to their favorite large language model, asking questions such as “What’s the most fun you can have in Coronado, with all your clothes on?” But, you can also use it to do tedious tasks that need to be carried out repeatedly — provided it knows how. That’s why I found myself teaching Claude a skill.

As someone who regularly writes articles about photography, I follow a standard workflow for handling the images I use. This includes writing a caption itemizing relevant metadata: camera, lens, aperture, shutter speed, and ISO. I also attach a  brief description of the photograph when adding it to the media repository. “Man biting dog in front of fountain” would be an example — and a photo worth seeing.

Each of these individual steps is tedious. And when you have to do it for twenty images in a typical 2,000-word article, it becomes incredibly tedious!

So, with the help of a collaborator, I began teaching Claude a skill, so that it could take care of it for me. The skill is called ‘Image Analysis’.

What is a skill?

In the world of Claude, the term ‘skill’ has a precise meaning. It is a folder of information that sits independently of Claude — the entity with whom I chat. The folder has a name (the skill), and contains a short description, a detailed prompt, reference material, templates, and even executable scripts.

Claude accesses the folder when I ask it to conduct a task requiring that specific skill. It then uses the contents of the folder to do the job I have set it. 

For example, I can upload a set of images as JPEG files, then ask Claude to prep them for publication using the Image Analysis skill. Claude duly conducts the analysis and produces a table of relevant information. Let’s take a look at how that works in practice.

A three-image batch

I uploaded to Claude the three images I subsequently sprinkled throughout this post. The table Claude produced, containing the specified information about them, is shown here.

File Name	Date	Caption	Alt Text
P1006371.jpg	14 June 2026	Panasonic DC-S5, Viltrox 28mm f/4.5, 1/320s at f/4.5, ISO 125. © Keith James	Two guitarists performing under an outdoor bandstand — one on bass guitar in a grey vest, the other playing a sunburst Gibson Les Paul, wearing a black hat and sunglasses.
P1006373.jpg	14 June 2026	Panasonic DC-S5, Viltrox 28mm f/4.5, 1/320s at f/4.5, ISO 125. © Keith James	A singer in a white shirt belts into a microphone on an outdoor stage, with a bassist performing behind him.
P1006380.jpg	14 June 2026	Panasonic DC-S5, Viltrox 28mm f/4.5, 1/320s at f/4.5, ISO 250. © Keith James	A smiling couple dance arm-in-arm on the grass at an outdoor concert.

It has four columns: file name, date captured, caption, and Alt Text (the short description I paste into the media repository).

Extracting the file name is pretty straightforward, as is the date of the image.

The clever bits are the caption and the Alt Text. For the caption, Claude extracts the required information from the EXIF data associated with the image, converting it into a standard format used throughout the photography world.

The Alt Text description is generated from the features Claude recognizes in the image. In my view, these descriptions are remarkable. They are concise, accurate, and informative. In the shot of the two musicians, Claude even points out that the guitarist is playing a sunburst Gibson Les Paul. Wow!

I am sure you can recognize the incredible productivity boost I gain from delegating to Claude responsibility for creating this table. And, Claude does a much better job than I would with the Alt Text descriptions.

I can use my time for much more creative tasks, such as making a cup of tea, or taking more photographs.

And this is just the beginning. There are many other tedious steps involved in creating a post in WordPress — the platform we use to publish the very article you are reading.

I won’t go into all the gory details here. But for a well-organized post on a well-organized blog-site, we need to add relevant internal and external links, assign categories and tags, choose a key-focus-phrase, and write excerpts and metadescriptions.

Those are quite a few steps, and many, if not all, can be automated by teaching Claude the relevant new skills.

I have posted several articles over recent months highlighting the risks associated with powerful LLMs. These are real, and deserve to be taken seriously. But it’s easy to overlook the potentially enormous benefits AI systems can bring in productivity improvements and problem solving.

Hopefully you can see that this simple example of a labor-saving skill I have described, if replicated across millions of tedious tasks, can be a powerful source of increased productivity and improved job satisfaction.

Is there a tedious job you regularly have to carry out, which you could delegate to your favorite LLM? The one I use, Claude, is a fast learner and eager to help out — it’s an all round star pupil!