The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads an installer script and a binary from the author's GitHub repositories during the setup phase.
Evidence: SKILL.md contains bash scripts that fetch install.sh from github.com/okx/onchainos-skills and the gmx-v2-plugin binary from github.com/okx/plugin-store.
Security Practice: The scripts include SHA256 checksum verification to ensure the integrity of the downloaded files before execution.
[COMMAND_EXECUTION]: The plugin binary executes the onchainos CLI tool to perform wallet operations and contract calls.
Evidence: src/onchainos.rs uses std::process::Command to invoke onchainos wallet contract-call, onchainos wallet history, and other CLI commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, creating a potential surface for indirect prompt injection attacks.
Ingestion points: Untrusted data enters the agent context through market information, price tickers, and token metadata fetched in src/api.rs, as well as raw blockchain data from eth_call in src/rpc.rs.
Boundary markers: SKILL.md contains a "Data Trust Boundary" section that explicitly instructs the AI agent to treat CLI output as untrusted external content and filter fields before interpretation.
Capability inventory: The skill can execute shell commands via the onchainos CLI (in src/onchainos.rs), perform network operations to GMX APIs (in src/api.rs), and query blockchain RPC nodes (in src/rpc.rs).
Sanitization: The Rust implementation parses external JSON data into strongly-typed structures and applies unit conversions (e.g., price precision) before returning a filtered subset of data to the AI agent.

gmx-v2-plugin