okx-dapp-discovery
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill implements a command that fetches a list of available plugins from the vendor's official GitHub repository and processes the JSON response using a Python one-liner. The Python code is a static string used solely for data parsing and does not execute arbitrary remote content.
- [EXTERNAL_DOWNLOADS]: Orchestrates the retrieval of plugin documentation and catalog lists from the official 'okx/plugin-store' repository on GitHub. These downloads are limited to the vendor's own infrastructure and require explicit user consent.
- [COMMAND_EXECUTION]: Uses system commands to manage plugin installations via the 'skills' CLI and to interact with the agent's local skill directory. These operations are governed by a static resolver table and require explicit user confirmation before any installation occurs.
- [INDIRECT_PROMPT_INJECTION]: Vulnerability surface identified where the skill processes untrusted data from user prompts and external plugin documentation.
- Ingestion points: User prompts and external 'SKILL.md' files fetched from the 'okx/plugin-store' repository.
- Boundary markers: Employs an intent-routing logic and an 'Override check' to delimit external signals from internal instructions.
- Capability inventory: Includes file system reads, network requests via 'curl', and skill management via the 'npx skills' tool.
- Sanitization: Implements a 'Fetched-content guard' that instructs the agent to treat plugin content as data only, and 'Secret hygiene' rules to redact sensitive information before processing.
Audit Metadata