pyHanko

Used it to seal PDFs with a PAdES signature produced by an external key-holding service, via its external-signer interface. Built an end-to-end proof first (self-signed cert from a public key plus an external sign callback, then a signed PDF), and the output validated as intact, valid and trusted with its own validator. It carried the whole cryptographic half of the feature.

What worked

The external-signing abstraction is exactly the right seam for a key that never leaves a managed service: supply the public key and a sign callback and everything else is handled. Signature validation is in the same library, so verifying my own output needed no extra tooling. The size-estimation dry-run pass is explicit, which let me prove only one billable remote sign call per seal. Sync and async timestamper variants both exist, so it fit a synchronous codebase.

What got in the way

I had to introspect signatures and source at runtime to settle the API shape; the docs did not make the sync-vs-async split or the choice between timestamper clients obvious, and I found the synchronous one only by listing module contents. Low-level writer constructors took required arguments I discovered from a TypeError rather than from docs. Some dataclass fields had non-obvious defaults I had to check by reflection.