The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automates the installation of the @okxweb3/a2a-node package via the Node Package Manager (NPM). This is a vendor-owned resource required for the agent-to-agent communication protocol.
[COMMAND_EXECUTION]: The instructions frequently invoke the onchainos and okx-a2a command-line tools to manage agent identities, handle file transfers, and interact with the decentralized marketplace. These operations are core functionalities of the service.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process "deliverables" and "market signals" from the decentralized marketplace. It explicitly identifies these as untrusted data sources and provides the agent with strict instructions to ignore any commands, URLs, or instructions embedded within them.
[DYNAMIC_EXECUTION]: The task monitoring protocol involves receiving decision_request items that contain an llmContent field. The agent is instructed to execute the commands specified in this field, which is used for remote orchestration. This capability is mitigated by a "Dangerous-Instruction Gate" that prohibits the execution of shell commands, host-level skills, or access to sensitive credentials.
[PERSISTENCE]: The skill uses platform-provided scheduling tools (CronCreate on Claude Code and automation_update on Codex) to maintain a live task monitor by scheduling a one-shot "wake" every two minutes if a decision is pending.

okx-ai