feat(node): accept a wallet's personal_sign signature on a transaction

[MehranMazhar]

What this changes

The node now accepts a wallet's personal_sign signature on a transaction, next to the signature it accepted before.

MetaMask and Trust Wallet will not sign a bare hash. They sign a text, and they hash it with a fixed prefix (EIP-191). Until now the node checked only a signature over the hash string itself, so no wallet could send a transaction.

Transaction::verify_signature tries two schemes. Each one is checked against its own digest, so one cannot pass for the other:

Scheme	The key signs	Used by
1 (unchanged)	the hash string	the SDK's local key, the mint authority, the faucet
2 (new)	clutch-tx:{chain_id}:{hash} with personal_sign	a wallet

In scheme 2 the hash is 64 lowercase hex characters with no 0x. The text names the chain, and verify_hash already ties the hash to chain_id, so a signature cannot move to another chain or to another transaction.

The wire format, the hash, the block signature and the mint cosignatures do not change.

This is a consensus change

A node without this change rejects a block that carries a scheme 2 transaction. Every validator must run this build before the first wallet transaction is sent. Plan: stage gets it first (the 3 stage nodes move together); mainnet gets it through a chain reset, because no CLT has ever been minted there.

Tests

• personal_sign_bytes_has_the_eip191_layout
• personal_sign_digest_matches_the_published_hello_world_vector (the digest comes from the ethers documentation, not from this code)
• a_signature_made_by_a_javascript_library_verifies (made by @noble/secp256k1, which the SDK uses)
• a_wallet_signature_binds_the_signer_and_the_exact_text
• in transaction.rs: wallet_signing_text_names_the_chain_and_the_bare_hash, verify_signature_accepts_a_wallet_signature, verify_signature_accepts_the_sign_personal_helper, verify_signature_still_accepts_a_signature_over_the_hash_string, verify_signature_accepts_the_signature_a_javascript_library_made, and five refusals (another chain, another key, another transaction, a flipped recovery id, a malformed signature)
• tests/wallet_signature.rs: a wallet-signed transfer goes through the pool and an authored block and is applied; two bad signatures are refused

Nothing was run on the author's machine. The Test workflow is the check.

Also

CLAUDE.md says the new scheme in "Gotchas", and its old line "There is no CI job running cargo test" is corrected: test.yml has run on every PR for some time.

🤖 Generated with Claude Code