air-agentic-wallet

Pass

Audited by Gen Agent Trust Hub on Mar 25, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides several Node.js scripts (e.g., air-send.mjs, air-approve.mjs, air-execute.mjs) designed for wallet operations. The agent is instructed to execute these scripts using the node command to facilitate transactions and signing.\n- [DATA_EXFILTRATION]: To perform its primary function, the skill reads sensitive cryptographic keys from the local file system (e.g., p256-private-key.pem) to generate local signatures. These signatures, along with wallet identifiers (e.g., userId, walletId), are transmitted to an external signing API (airApiAgentSignUrl) and blockchain RPC/bundler endpoints. This behavior is consistent with the intended purpose of the wallet skill.\n- [EXTERNAL_DOWNLOADS]: The skill depends on the viem library and interacts with external blockchain infrastructure and services, including the AIR signing API, Privy RPCs, and third-party bundlers like Candide.\n- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface because it processes untrusted data from external sources.\n
  • Ingestion points: External data enters the agent context through the JSON handoff bundle and various user-provided transaction parameters.\n
  • Boundary markers: No specific delimiters or safety instructions are used to wrap or sanitize the external data within the prompts.\n
  • Capability inventory: The skill possesses the ability to make network requests, read and write local configuration files, and execute cryptographic signing operations for on-chain execution.\n
  • Sanitization: While the underlying scripts use the viem library to validate Ethereum addresses and hexadecimal data formats, the agent remains responsible for the secure passing of arguments to the scripts.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 25, 2026, 02:30 AM