Hi Devs,
Two small transaction introspection opcodes that are worth considering are OP_PUSH_KEY_INTERNAL or OP_PUSH_KEY_EXTERNAL which can return the taproot key for the current input.
While the internal key could be included in the tree already, and this is just a performance improvement, the external key creates a hash cycle and is not possible to include directly.
This came up as a potential nicety while looking at how BIP-118 "puns" a single 0x01 byte as a key argument to refer to the Internal key for compactness. It would be more general if instead of 0x01, there were an opcode that actually put the Internal key on the stack.
There is a small incompatibility with BIP-118 with this approach, which is that keys are not tagged for APO-enablement. Thus, there should either be a version of this opcode for APO tagged or not, or, APO should instead define some CheckSig2 which has APO if tagging is still desired. (Or we could abandon tagging keys too...)
It might be worth pursuing simplifying APO to use these OP_PUSH_KEY opcodes because future plans for more generalized covenant might benefit from being able to get the current key off the stack. For example, TLUV might be able to be decomposed into simpler (RISC) opcodes for getting the internal key, getting the current merkel path, and then manipulating it, then tweaking the internal key.
The internal key might be useful for signing in a path not just for APO, but also because you might want to sign e.g. a transaction that is contingent on a HTLC scriptcode being satisfied. Because it is cheaper to use the 0x01 CHECKSIG than doing a separate key (<pk> CHECKSIG), it also causes an unintended side effect from APO of incentivizing not using a unique key per branch (privacy loss) and incentivizing enabling an APO tagged key where one is not required (unless 0x00, as I've noted elsewhere is added to the 118 spec as a pun for an untagged key).
Pushing the external key's use is less obvious, but with the development of future opcodes it would be helpful for some recursive covenants.
Both opcodes are very design specific -- there's only one choice of what data they could push.
Of course, we could keep 118 spec'd as is, and add these PUSH_KEYs later if ever desired redundantly with the Checksig puns.
Cheers,
Jeremy