Developer

Everyday programming utilities. Test a regular expression against sample text and see the matches, read a cron expression in plain English before trusting it to a scheduler, decode a JWT to see what a token actually claims, write HTML, CSS and JavaScript with a live preview, and generate realistic test data that is safe to publish.

Each of these is a place where a small misreading becomes a production incident: a regex that backtracks catastrophically on real input, a cron expression whose two day fields are ORed rather than ANDed, a token trusted without checking its signature. The tools here show you the actual behaviour — the matches, the next run times, the decoded claims — rather than asking you to trust a description of it.

The same applies to a diagram. A .bpmn file carries both the process and the coordinates of every shape in it, so it can be drawn exactly as its author placed it rather than re-laid-out into something else — alongside an inventory of every element and the modelling problems the file contains.

An API description is the same kind of file. An OpenAPI or Swagger document, pasted as YAML or JSON, reads as the list of operations it describes — each with its parameters, request body, responses and a curl command to try — together with the places it breaks the specification: a path parameter that is never declared, a reference that points at nothing, an operationId used twice.

Starting a repository comes with two files nobody writes from memory: a .gitignore, put together from GitHub’s own templates for your languages, editors and operating system, with a path tester that says which line ignores a file and why; and a licence, the standard text with your name and the year where it has a place for them, and the current SPDX identifier for the top of each file.

Then there are the Git commands themselves, found by what you want to do — undo a commit, rename a branch, find the commit that introduced a bug — each marked as safe, as rewriting history or as able to lose work, with your own branch and file names filled in once you type them.

Prompts are counted in tokens rather than characters, and the ratio between the two shifts with language and with code. A token counter gives the exact figure GPT-5, GPT-4o and GPT-4 would see, with each token drawn as its own block so the reason a prompt is expensive — a long identifier cut into pieces, a script spent a token per character — is there to read.

A page’s markup can be checked for what a screen reader depends on: its heading outline and landmarks, and the images, links, buttons and fields that have nothing to read out, each finding graded and linked to the WCAG criterion it relates to where there is one.

And when a file will not say what it is, its bytes will: a hex view of any file, its real type named from the signature in its first bytes, the numbers at any offset read as integers, floats or a timestamp, and dumps that match xxd and hexdump byte for byte.