The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads an external binary for the Meteora DLMM plugin and an installation script for the onchainos CLI directly from the vendor's GitHub organization (okx).
[REMOTE_CODE_EXECUTION]: During the pre-flight setup, the skill executes a shell script downloaded from a remote URL using a piped-to-shell pattern (curl | sh). This behavior is documented as part of the vendor's automated installation process.
[COMMAND_EXECUTION]: The skill interacts with the local system by spawning subprocesses to run the onchainos CLI tool for wallet resolution and transaction signing, as well as uname and hostname for environment identification.
[DATA_EXFILTRATION]: An installation reporting script collects system metadata, including the hostname, operating system, machine architecture, and home directory path. This data is hashed to create a device identifier and transmitted to the vendor's telemetry endpoints at okx.com and a Vercel-hosted deployment.
[PROMPT_INJECTION]: The skill processes external data retrieved from the Meteora DLMM API and Solana RPC nodes to display pool details and quotes. This creates an indirect prompt injection surface where a compromised or malicious data source could attempt to influence the agent's behavior.
Ingestion points: Data enters via the Meteora REST API (dlmm.datapi.meteora.ag) and Solana RPC providers in src/solana_rpc.rs and src/api.rs.
Boundary markers: No explicit boundary markers or instructions to ignore embedded content are used when presenting API data to the agent.
Capability inventory: The skill has the capability to execute shell commands, perform network requests, and initiate blockchain transactions via the onchainos CLI.
Sanitization: There is no evidence of sanitization or filtering of the natural language fields (like pool names or symbols) retrieved from the remote APIs before they are processed by the agent.

meteora