Extract Code Blocks

Extracting code blocks is the inverse of stripping Markdown: keep what is inside the fences and discard everything around it. The value is in the second half, because the prose around code in a chat transcript or a documentation page is usually longer than the code itself and entirely unwanted when you are pasting into an editor.

Fenced blocks, the ones delimited by triple backticks, are unambiguous and extract reliably. Many carry a language hint on the opening fence, which is worth keeping if you intend to save each block to a correctly named file. Indented code blocks, the older convention of four leading spaces, are harder: four spaces is also how people indent a nested list, so a purely indentation-based rule will occasionally extract prose and occasionally miss code.

Indentation inside the block is the thing that must survive untouched, and it is the thing casual copying most often breaks. In Python it is syntax rather than style, so a lost level is a program that no longer runs. Tabs converted to spaces, or spaces normalised to a different width, change files in ways a diff will show and a reviewer will not thank you for. Extraction should copy the block byte for byte and do nothing clever.

Multiple blocks raise a question worth answering before you start: do you want them concatenated into one file, or kept separate. A chat that built a program step by step usually wants them in order as one file. A page documenting several independent examples usually wants them apart, since running them together makes no sense.

One thing to check by eye: chat interfaces sometimes insert invisible characters or convert straight quotes to curly ones inside what they display as code. Curly quotes are not valid string delimiters in any language, and the resulting syntax error points at a line that looks perfectly correct. If extracted code fails to parse and the line looks fine, that is the first thing to suspect.

The extraction runs in your browser, which is the point when the transcript you are pulling from contains an API key, an internal hostname or a customer identifier in an example, as transcripts routinely do.