x402-erc7710
Fail
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The instructions direct the agent to generate and export private keys for 'agent wallets' in JSON format (e.g., in the 'Browser Lab Pattern' and 'Conversational Operator Mode'). Handling and exporting raw private keys, even for testnets, is a dangerous practice that increases the risk of credential theft and exposure.\n- [PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection because it parses and acts upon data received from external, untrusted 'Resource Servers' via HTTP 402 responses.\n
- Ingestion points: The agent reads network, token, amount, and payee information from external HTTP headers and response bodies (SKILL.md).\n
- Boundary markers: The skill suggests implementing an
AgentSpendPolicyto check requirements, but these are instructions for the agent rather than hardened system boundaries.\n - Capability inventory: The agent has the capability to execute on-chain transactions, spend ERC-20 tokens, and request wallet permissions (SKILL.md, references/demo-blueprints.md).\n
- Sanitization: The agent is instructed to manually verify requirements against a policy before making payments.\n- [COMMAND_EXECUTION]: The skill requires the agent to generate and execute blockchain-related scripts and transaction payloads, including preparing ERC-7715 permission requests and applying revocation decisions through local EVM controller paths (SKILL.md).
Recommendations
- AI detected serious security threats
Audit Metadata