#46507 suggests multiple strategies which make use of automatically generated dictionaries of interesting values. It is likely also valuable to allow users to generate their own dictionaries of interesting values, as is supported by AFL and libFuzzer, that the mutation engine can pull from.

The easiest solution for encoding the values is likely to use a similar encoding scheme to AFL and libFuzzer, so existing dictionaries can be easily reused. Both use quoted strings with \x byte encoding, e.g.:

"blah"
"\"ac\\dc\""
"\xF7\xF8"
"foo\x0Abar"

AFL and libFuzzer support prefixing these values with a keyword (i.e. name="blah"), but do not use the name for anything. It seems reasonable to support this, so dictionaries can be directly imported, but to just ignore these values. Lines prefixed with # are ignored.

These dictionaries could be consumed from the testdata folder, similarly to the testdata/corpus seed inputs, and by extending the testing.F API with something like func (f *F) AddDictionaryEntry([]byte).