evm-rpc
Audited by Snyk on Nov 19, 2026
MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).
- Third-party content exposure detected (low risk: 0.10). The required workflow reads Ethereum/EVM chain state (balances, blocks, eth_call results, receipts) only for addresses or hashes the agent or user selects, with no outsider-writable feed, so outsiders can publish data but cannot force the agent to read it.
MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).
- Direct money access detected (high risk: 0.70). The skill is mostly an RPC integration guide: balance reads, block queries and ERC-20 reads are read-only information. Cycle costs are service costs for RPC calls and are excluded. The one financial-scope item is the
send_raw_transactionupdate method (lines 242-250), which broadcasts a pre-signed EVM transaction to the network. That submits a financial instruction (a transfer or other transaction) for downstream execution, so it is a direct operation. The pitfall at line 80 says the canister does not sign and that the caller must sign, but it does not describe any access control.
The exported #[update] method has no caller check, no authentication and no authorization. It accepts any signed_tx_hex from any caller. The only gate is that a valid signature must exist. That is a local signing/key-presence check, which does not count as independent access validation. No documented service-enforced authentication or approval covers this route. The relay is generic, though: it does not itself build or sign transfers, and it is partly an implementation example for building canisters. Even so, the skill offers a working route for submitting signed transactions without a qualifying control. That puts the route at high severity.
Issues (2)
Third-party content exposure detected (indirect prompt injection risk).
Direct money access capability detected (payment gateways, crypto, banking).