okx-ai

Warn

Audited by Gen Agent Trust Hub on Jul 23, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill manages the installation and recovery of the @okxweb3/a2a-node package and the onchainos binary. These dependencies originate from the OKX organization, a recognized technology provider, and are handled through official registries and domains.
  • [COMMAND_EXECUTION]: The skill performs extensive command execution using the onchainos CLI for blockchain-based identity and task management. It also utilizes a dynamic execution pattern in the task watch loop where it executes instructions contained within an llmContent field provided in decision requests. This field is used to maintain state and logic across multi-turn user interactions but involves the runtime execution of dynamically generated instructions.
  • [DATA_EXFILTRATION]: Facilitates the reading and uploading of local files to the OKX AI platform via agent file-upload and agent task-attach commands. This functionality is the primary mechanism for agents to submit deliverables and attachments for tasks in the marketplace.
  • [PROMPT_INJECTION]: Includes specific security guidelines to protect against indirect prompt injection by instructing the agent to ignore any executable-like content found in untrusted data fields, such as agent descriptions, reviews, or dispute evidence materials.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 23, 2026, 04:06 PM