The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs users to install the okx-outcomes binary by piping a script from the official GitHub repository of the vendor (https://raw.githubusercontent.com/okx/outcomes-cli/main/install.sh) to the shell. This is the documented installation method for the CLI tool.\n- [EXTERNAL_DOWNLOADS]: The skill requires the installation of the @okx_ai/okx-trade-cli Node.js package from the NPM registry, which is the official wrapper provided by the vendor for this skill's functionality.\n- [COMMAND_EXECUTION]: The skill executes commands using the okx outcomes binary. It provides detailed documentation for command routing and handles binary execution within the agent's environment safely by inherited stdio.\n- [CREDENTIALS_UNSAFE]: The skill references the PREDICTIONS_AGENT_PRIVATE_KEY environment variable for transaction signing but includes proactive safety instructions. It explicitly warns against sharing keys in chat, recommends the OS keyring for storage, and defines a secure EOA binding process.\n- [PROMPT_INJECTION]: The skill processes external market data that could serve as a vector for indirect prompt injection.\n
Ingestion points: Data is ingested through multiple okx outcomes data commands as defined in the skill documentation.\n
Boundary markers: The skill enforces the use of --json for structured data and sets clear instruction boundaries in the workflow.\n
Capability inventory: The skill allows for high-risk write operations including order placement and token redemption.\n
Sanitization: All high-risk operations require a mandatory dry-run preview followed by an explicit human confirmation (replying "confirm"), which effectively mitigates accidental or malicious execution triggered by external data.

okx-outcomes