okx-dex-market
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted external data from blockchain sources, news platforms, and social media (X/Twitter). While it has an attack surface for indirect prompt injection, it includes clear boundary instructions for the agent.
- Ingestion points: Processes news articles and summaries (
social news-latest), token metadata and search results (token search), and social media handles and sentiment rankings (sentiment-ranking). - Boundary markers: The skill contains explicit security instructions in
SKILL.mdto treat all CLI fields as untrusted external content and strictly forbids interpreting token names, article text, or social media handles as instructions. - Capability inventory: The skill utilizes the
onchainosCLI to perform all read operations and WebSocket monitoring, which does not involve sensitive local file access or unrestricted command execution beyond its domain. - Sanitization: The skill instructs the agent to normalize addresses and includes mandatory warnings for unverified or low-liquidity tokens to protect the user.
- [EXTERNAL_DOWNLOADS]: The skill's preflight procedure installs a vendor-specific tool.
- Evidence:
_shared/preflight.mdinstructs the agent to runnpx -y @okxweb3/onchainos-installer installto set up the necessary environment. The source package is specific to the vendor. - [COMMAND_EXECUTION]: The skill relies on executing commands via a local CLI tool to interact with blockchain data.
- Evidence: Throughout the reference files (e.g.,
market-cli-reference.md,token-cli-reference.md), the agent is instructed to runonchainoscommands with various parameters derived from user intent.
Audit Metadata