Tools are defined via AgentKit's MCP bridge (getMcpTools), not directly visible in source. Tool definitions are inferred from tool names and brief descriptions only. No explicit input schemas, parameter descriptions, or output schemas are visible in the provided code. Descriptions are present but minimal (10-60 chars), lacking context on WHEN to use each tool, prerequisites, or return structure.

• Expose tool definitions directly in source code (not via AgentKit bridge) with explicit JSON Schema input/output definitions. Include parameter types, constraints (enums, min/max), and descriptions for every parameter.
• Expand tool descriptions to 50 - 200 characters. For each tool, answer: What does it do? When should the LLM call it? What does it return? E.g., 'Fetch live token prices from Pyth price feeds. Accepts token symbol (e.g., ETH, BTC) or Pyth feed ID. Returns price in USD and feed metadata. Use before converting amounts or checking portfolio value.'
• Add parameter descriptions for all inputs. E.g., for wallet_send_native_token: 'recipient (string, required): Recipient wallet address (0x...). amount (number, required): Amount in native tokens (e.g., 1.5 for 1.5 ETH). Must not exceed wallet balance.'
• Implement dry-run and confirmation patterns for irreversible operations. Add a 'dry_run' boolean parameter to wallet_send_native_token and erc20_transfer. When true, return a preflight summary (recipient, amount, estimated gas) without executing. Require explicit confirmation before actual transfer.
• Document error handling and recovery paths. For each tool, specify: What errors can occur? Are they retryable? What should the LLM do next? E.g., 'If balance is insufficient, return error 'INSUFFICIENT_BALANCE' with available balance. LLM should ask user to reduce amount or fund wallet.'