skills/bankrbot/skills/berry-juicer/Gen Agent Trust Hub

berry-juicer

Pass

Audited by Gen Agent Trust Hub on Jul 5, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides shell script examples utilizing standard utilities such as curl, jq, openssl, and cast for API interaction and cryptographic signing. These commands are executed locally to facilitate agent automation.
  • [EXTERNAL_DOWNLOADS]: Authenticated requests are made to juicerapi.berryfi.org and api.bankr.bot. These communications are restricted to necessary protocol configuration and wallet operations.
  • [PROMPT_INJECTION]: The skill explicitly addresses the risk of indirect prompt injection by instructing the agent to ignore operational parameters (such as addresses, amounts, or contract targets) if they originate from untrusted model outputs or external data. It mandates the use of pinned allowlists and human confirmation for all on-chain actions.
  • [DATA_EXFILTRATION]: Authentication is performed using short-lived Ethereum personal_sign signatures instead of static API keys. The 'v2' signing implementation binds signatures to specific request parameters (host, method, path, body hash), which effectively prevents replay attacks and credential theft.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 5, 2026, 02:11 PM
Security Audit — agent-trust-hub — berry-juicer