Cross-chain swaps in one line — built for AI agents. Auto-routing, bridge, gas, slippage protection.
AutoSwap defines 2 tools with complete input schemas and enums for chain selection. Both tools have descriptions (swap: 56 chars, quote: 50 chars) meeting the 10-1024 char baseline. However, descriptions lack WHEN/WHY context and actionable guidance. Parameters are well-typed with descriptions, but lack dependency documentation (e.g., wallet_key security implications, dry_run behavior). Output schemas are not documented in the source. Error handling and recovery guidance are absent. No tool annotations (readOnlyHint/destructiveHint) despite swap being IRREVERSIBLE. The swap tool exposes wallet_key as a parameter, a critical security anti-pattern.
Get a price quote without executing the swap.
Execute or simulate a cross-chain token swap.
wallet_key exposed as tool parameter, credentials must never appear in parameters. Use server-side secret injection via environment or vault. Agent traces log all parameters, leaking secrets into logs and prompt history.
swap tool marked IRREVERSIBLE but lacks tool annotations (destructiveHint). LLMs cannot infer irreversibility from risk labels alone. Add destructiveHint annotation and require explicit confirmation or dry_run validation before execution.
Output schemas not documented. LLMs cannot plan downstream calls or extract required fields (e.g., transaction_hash, actual_slippage, gas_cost). Document return types for both swap and quote tools.
Descriptions lack WHEN/WHY context and recovery guidance. 'Execute or simulate a cross-chain token swap' does not explain when to use swap vs quote, what dry_run does, or how to handle slippage errors. Expand to 100-150 chars with actionable context.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
No error handling or recovery guidance. If swap fails due to insufficient liquidity, slippage exceeded, or network error, the LLM has no guidance on retry strategy, fallback tools, or user-fixable actions.