feat: Hardware Key Agent - command hint

[Joerger]

Part of RFD 199

Includes the client command in hardware key agent server prompts to provide context.

Also fixes the proxy host context in Teleport Connect hardware key prompts, which actually regressed in #53703 due to the proxy host being parsed by routing.parseClusterName into "" on the front end.

#54026

[zmb3]

Any security concerns with allowing the client to provide the command hint? Could a malicious actor abuse this to trick a user into entering their PIN to perform a signature?

[Joerger]

Any security concerns with allowing the client to provide the command hint? Could a malicious actor abuse this to trick a user into entering their PIN to perform a signature?

It is definitely a valid concern, but one that we ultimately decided to accept in the RFD phase:

• RFD 199 - Teleport Key Agent #51536 (review)

  The most important change introduced with these RFDs is the gRPC service, exposed via a UNIX domain socket. The fact that the socket is only available locally, restricts the threat model to an attacker that already has access to, or is able to execute commands on the victim's machine. This is generally considered a privileged position.

• RFD 199 - Hardware Key Agent w/ PIN Caching #52495 (comment)

  it's already possible for a third party client to claim a PIV transaction with [touch cached for 15 seconds or] the PIN cached on the card, within the ~5 second grace period.

Any type of attack to misuse the hardware key agent requires the attacker to have remote access to the user's machine, which is a privileged position on its own. At this stage, some attacks are possible to coax the user into tapping or entering PIN on their YubiKey, but they all require action on the part of the victim to complete.

If you think we need to make the text here more "red", like we do with headless, we can explore that. But keep in mind that the agent will also be used by some Teleport Connect internal functionality which call tsh commands under the hood, like connecting to servers which just calls tsh ssh.

Headless prompt:

FWIW the command hint here is not meant as a security mitigation, rather it's intended as a UX improvement so users are not confused as to the origin of prompts that they themselves initiate through other clients. e.g. the first time a user runs tsh ssh and gets prompted for touch/PIN in Connect, we want to make it clear why.