Checksums, tokens, cron, encodings.

The part of a developer's day that is not thinking: decode this token, what does this cron line actually do, why does this string not match. All of it exact, none of it something a model should do from memory.

Start for freeSee the plans

Three steps.

  1. 1

    Paste the thing

    Token, cron line, string, file - whatever is not behaving.

  2. 2

    It calls a tool

    Not memory. A checksum from memory is a checksum that looks right.

  3. 3

    You get the exact answer

    With the reason, and with what was not checked.

A JWT decoded into its claims, with the unchecked signature stated and an invisible character found in the issuer.
In the app. A JWT decoded into its claims, with the unchecked signature stated and an invisible character found in the issuer.

What it does.

  • Checksums of a text or a file, and encodings both ways - base64, hex, percent, HTML
  • Decodes a JWT and says outright that the signature was not checked
  • Explains a cron expression and lists the next real dates, including the OR trap
  • Finds invisible characters, wrong line endings and lookalike letters
  • Validates JSON, YAML and XML with the line number, and runs XPath
  • Real random passwords and tokens from the operating system, with the entropy in bits

Where it stops

It does not run your code. Tools that touch a repository run on your own machine through the desktop app, where you see every command before it happens.

Asked like this

This JWT is being rejected by the gateway. Decode it, tell me whether it has expired, and check whether the issuer string has anything invisible in it.

Which size to use

Small for all of it. These are tool calls; the model only has to pick the right one.

Compare the sizes

Questions.

Is my token sent anywhere?

It is decoded on our server and not stored beyond the conversation. The signature is never verified, so no key is needed or wanted.

Can it run commands for me?

In the desktop app, on your machine, with every command shown before it runs. Never on our server.

Why not just ask the model?

Because it will answer. A model produces a checksum-shaped string with the confidence of a real one, and that is the failure mode worth avoiding.

Other kinds of work.

Coding

Reads the repository, runs the build, reads the errors.

Reasoning

Works a problem through, and says where the reasoning stops.

Office work

Reads what piled up, writes what has to be written.

All kinds of work