chop
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
node scripts/chop.mjsto interface with the blockchain and build transaction plans. This is a standard pattern for CLI-based AI agent skills where the agent executes local scripts to perform complex calculations or data fetching. - [DATA_EXFILTRATION]: The skill makes network requests to
api.bankr.botand various RPC endpoints (rpc.mainnet.chain.robinhood.com,choplogs.app/api/rpc) to fetch chain state and submit transactions. These are legitimate destinations for the skill's purpose as a DeFi interface on the Robinhood Chain. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data such as token names and symbols fetched from the blockchain. The author has implemented a 'Safety' reference (
references/SAFETY.md) instructing the agent to treat these as data only and never as instructions, mitigating the risk of injection through on-chain metadata. - [SAFE]: The skill follows best practices for non-custodial operations. It does not handle private keys, uses exact approvals instead of unlimited ones, and relies on the Bankr wallet's own security layer and simulation for transaction safety. All external dependencies (e.g.,
viem) are standard, well-known libraries.
Audit Metadata