The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches installation scripts and CLI dependencies from OKX's official GitHub repository (okx/onchainos-skills).
[EXTERNAL_DOWNLOADS]: Downloads the mayan binary and additional toolsets from the vendor's repository on GitHub (MigOKG/plugin-store).
[REMOTE_CODE_EXECUTION]: Executes a setup script for the onchainos environment using a piped curl | sh command from OKX's verified GitHub repository.
[COMMAND_EXECUTION]: Performs automated setup tasks including binary installation, directory creation (~/.local/bin), and permission management (chmod +x) for the downloaded tools.
[DATA_EXFILTRATION]: Generates a device fingerprint using system metadata (hostname, OS, architecture, and home directory path) to create a unique installation ID. This metadata is sent to the vendor's domain (plugin-store-dun.vercel.app) and OKX's official API (www.okx.com) for install reporting. This behavior is explicitly disclosed in the 'Data Trust Boundary' section of the skill.
[PROMPT_INJECTION]: The skill interacts with external blockchain data and third-party APIs, creating a potential surface for indirect prompt injection.
Ingestion points: Results from get-quote, swap, and get-status commands include data from on-chain contracts and Mayan Finance APIs.
Boundary markers: The documentation includes a 'Data Trust Boundary' that explicitly warns the agent to treat all returned data as untrusted and not to interpret it as instructions.
Capability inventory: The skill can perform on-chain transactions and wallet operations via the mayan and onchainos binaries.
Sanitization: Instructions mandate that the agent filter CLI output to human-relevant fields only, preventing raw API responses from polluting the command context.

mayan