This cost 20% of the daily plan limit with Claude

9 November 2025

You know like, I be vibing.

I reached my limit for the day using Claude so here i am writing a blog until the limit resets again!

But what is a limit and how is it calculated? It's quite complicated as it works out.

There are many factors at play such as the size of the current context it has, the size of the code base it's updating, the complexity of the request and so on.

It's almost impossible to work out how much a feature will cost or how many tokens it took to execute your shiny new feature.

That being said, this is what cost me 20%.

Prompt wrangling

At the time, I had a one-thousand-line-long markdown file listing of the features and instructions of my app.

I have a context at the top outline various strategies that i want it to adhere to. Such examples would be to

• Use the shared http client for all api calls
• Save all DB migrations to database/migrations
• Ensure every endpoint has a permissions check
• etc etc

I usually start a fresh context every day or for a new feature where it doesn't need to know about any existing logic.

At the start I command the LLM to understand the context section before anything.

Then we are away and this is the minimum context i start with. No idea how much this costs?

The Prompt

Ok, i will stop beating around the bush.

I required a multi-file uploader and this is how i got it.

## Images component /account/details/images
-  create a child route under account/details/images for adding and updating images
  -  Only business owners can access this route
-  Add file upload component
  -  This should be a generic re-usable component
  -  Allow uploading multiple images
    -  Max 10
    -  Each image max size 1MB
    -  png, jpg, jpeg only
-  Image uploading
  -  Images should be uploaded one at a time with a progress bar
  -  parse the image name to only allow alphanumeric, dashes and underscores
  -  Save the images to /images/<company_id>/<file-name>.<png, jpg, jpeg>
  -  convert the image to thumbnail
    -  the size should be good for mobile view
    -  it should respect the orientation of the image
    -  Save the thumbnail to /images/<company_id>/thumbnails/<file-name>_thumb.<png, jpg, jpeg>
  -  Create a new table business_images
    -  it should have a uuid, company_id, file-name, thumbnail_path, created_at
    -  create a new migration file for this
  -  Save the info of image uploads to the database
-  The page should call a new endpoint /business/details/images
  -  GET /business/details/images - get all images for the company
  -  POST /business/details/images - upload a new image
  -  DELETE /business/details/images/:id - delete an image by id
-  When the component loads it should fetch the image list from the database
-  Show all the images in a grid format from the thumbnails
-  Each image should have a delete button
 -  When clicked display a confirm modal
 -  If confirmed delete the image from the database and delete the files from the server
 -  Display a toast message on success or error

The Output

There are quite a lot of file changed and updated here.

The Display

Note: this is a picture after i also added `'set a main image' and 'ordering of image' prompts. So all you would see is the blue choose file, red delete and the image list.

The End (Almost)

This prompt was more on the larger side to what i usually do but the result was good.

It almost worked fully. There was one tweak required to display the image from the server in the front end but a minor fix.

In total there are about nine new files and around the same number of edited files, so a big chunk to review.

Knowing the size of the context, number of token required is nearly impossible to predict.

Checking the usage before and after tallied up to be a whopping 20% of my daily plan - i'm on the cheapest paid plan - £20pm.

Am i happy with the result? Yes. It saved me a load of time and now have image uploads working. 'Working' meaning i have manually tested it, so that's good enough, ship it!

Multiply this 'effort' by 5, let's say 4 or even 3, as there are inevitability bugs so that could also cost 20-40%, not to mention the LLM writing some tests.

For the MVP-turn-production app, that's a pretty good output for one day.