Out of the box, check-spelling will scan everything. Whether that's a .png or .zip.

In general, users are requested to set up an exclude file.

It would be nice if the tool could recognize files that aren't interesting to it and suggest an exclude file, just as it can suggest things to add into the whitelist.

If a file has more non-words than dictionary words, it's a good candidate for exclusion.

This is implemented in v0.0.21 as part of Autodetect noisy files

1. Files would need to be scanned individually
2. Word counts would need to be maintained (currently the engine uses a hashset, it would have to switch to a hashtable where it would retrieve the value, increment, and store)
3. The fold phase(s) would need to merge the counts together (things like foo, Foo, and foos are merged into foo and mostly reported as a single entity)
4. When the dictionary pass is reached, as a word is found in the dictionary, its count would need to be added to a running total, and at the end, the count for the remaining tokens would need to be summed and compared
5. If there's an imbalance (51%?), a warning should be issued

This is implemented in v0.0.22 as part of Heuristic exclude suggestions

Totals for files could be checked by file extension, file infix, file prefix, and directory name

Probably with a different threshold (75%?), or different thresholds for each

Suggestions could be made based on each of those categories:

exclude.txt:

# .png files:
\.png$
# .min. infix
\.min\.
# jquery prefix
/jquery-
# test-data directory:
/test-data/