An nmh folder corresponds to a directory. There are no limits on folder names beyond those of the host filesystem.

The file name is a positive integer. Other files containing metadata or arbitrary names can exist in a folder; while the preference is that non-message files begin with “.”, all files that are not positive integers must be ignored by an MH-compatible implementation. However, implementations are free to indicate to the user the existence of non-message files that are not prefixed with a “.”.

The filename for a new message is one greater than the highest numbered message in the folder; its full path can be accessed by the pseudo-sequence new (e.g., mhpath new). New messages are only permitted to be added to a folder at the end of the message number range.

To add a new message to a folder, the recommended sequence is:

• Create a temporary file in the desired folder.
• Attempt to link the temporary file to the new message number.
• If successful, remove the temporary file. If the link fails, increment the message number and try again.

There is one context file. Its default location is in the user's Path and its default name is context, but these can be overridden by the $MHCONTEXT environment variable. context has the following format:

where folder is the directory name of the current folder. Lines beginning with “atr” are used for private sequences. sequence is the name of the private sequence, path is the full path to the folder with the private sequence, and m[-n] is a message number or range of message numbers in the sequence.

There is one sequences file in each nmh folder. Its default name is .mh_sequences, but this can be overridden by the “mh-sequences” profile entry. sequences has the following format:

showing the (possibly empty) message numbers and/or ranges of message numbers in each sequence. The cur sequence has at most just a single message number, not a range.

Sequence names have a maximum size of 998 characters. Each line is also limited to a maximum of 998 characters, but RFC 822 continuation rules apply; sequences can be continued across multiple lines by prefixing continuation lines with a whitespace character.

If an implementation finds messages in a sequence that do not exist, the sequence file should be updated to remove the missing messages from the sequence. If a sequence contains no messages, it should be removed from the sequence file. The exception to this is the cur sequence, which can refer to a nonexistent message.