Food for thought:

1. Externally-provided content is dangerous. You might use a hash of the domain name, but I'd avoid files named after the sources.

2. Metadata such as a date would be useful.

3. Despite 1, a highly-sanitised hostname could be informative. An iconv to 8-bit ASCII [-a-zA-Z0-9_], and not allowing the first character to be '-' might be a start. Put a length limit on that as well.