The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to install numerous official packages under the @goat-sdk ecosystem and references standard blockchain RPC endpoints for networks like Base, Arbitrum, and Solana. These represent normal developer resources for the skill's purpose.
[PROMPT_INJECTION]: As a toolkit for building financial agents, the skill establishes an attack surface for indirect prompt injection via user-supplied natural language instructions. However, the documentation proactively mitigates this by providing system prompt templates that require human confirmation for all transactions.
Ingestion points: User prompts processed by bridge-agent.ts, agent.ts, and langchain-agent.ts.
Boundary markers: System prompts in the examples explicitly instruct the agent to "Always confirm the action before executing transactions" and "Never send funds without the user specifying the exact amount."
Capability inventory: Onchain operations including token transfers, DEX swaps, and cross-chain bridging using the viem library and protocol plugins.
Sanitization: The toolkit uses Zod schemas in tool definitions (e.g., TokenInfoTools) to enforce strict validation of token addresses and parameters before they reach the blockchain layer.

goat