Single tool 'execute_skill_action' has a minimal schema with only one string parameter. Description is generic (67 chars) and lacks specificity about what 'deterministic, zero-dependency Agent Commerce skill operations' means. No parameter description provided for 'query_payload', LLMs cannot infer what format or content is expected. No output schema documented. No error handling guidance. Tool name uses verb but is vague ('execute_skill_action' vs 'buy_token', 'execute_purchase'). No indication of what side effects occur (payment processing is mentioned in client.py but not in tool definition). Critical security issue: tool accepts arbitrary 'query_payload' string with no validation or sanitization, yet performs financial transactions.

• Add a detailed description for 'query_payload' parameter: specify expected format (JSON object? string? enum?), required fields, and constraints. Example: 'JSON-encoded transaction request with fields: merchant_id (string), cart_items (array of {sku, qty, unit_price}), buyer_token (encrypted vault token).'
• Document the output schema explicitly. Specify all fields returned (status, message, order_id, total_charged_usd, settlement_status, receipt_url) with types and meanings.
• Expand tool description to 100-150 chars explaining the specific action: 'Execute a one-click instant buy transaction for a merchant. Processes payment, settles funds, and returns order confirmation and receipt URL. Irreversible, use with caution.'
• Implement a dry-run parameter (boolean, default false) to allow agents to preview the transaction before committing. Return estimated totals without charging.
• Add explicit error handling: return structured errors with recovery guidance (e.g., 'Invalid merchant_id. Call list_merchants() to see available merchants.' or 'Payment declined. Suggest user update payment method.') instead of generic success/failure.