The parse_ccs() Credential constructor takes a full CCS. RFC9528 Section 3.5.2 also describes that naked COSE keys can be used (and I think are common in ACE EDHOC profile) that are turned into a CCS by prefixing them with A108A101.

I suggest (and will probably later PR) to add a dress_cosekey() constructor that takes only a COSE key, extracts the key (possibly refactoring with parse_ccs) and adds the prefix. This is a minor usability help, but saves downstream crates from carrying a 4-byte magic byte sequence around.

Ideally, that parsing function should be less strict than what parse_ccs does now; in particular, it should support EC2 keys where only the X coordinate is given (as is sufficient for ECDH), and which have no kid is present. Is there any particular reason why this (and a subject) is required in parse_ccs(), other than that parse_ccs() was originally designed to ingest test vectors where those happened to be present?