Single tool with mixed quality. The tool name 'get_yield_pools' follows verb_noun convention, but the parameter descriptions lack constraint information (enums, ranges, format). The tool description is adequate (170 chars, within 10-1024 range), but parameter descriptions are minimal and do not specify valid formats or examples. Output schema is not formally documented, the function returns a List[Dict[str, Any]], which the LLM cannot introspect. Error handling exists (try/except with ctx.error) but lacks actionable recovery guidance. No pagination support despite the underlying API potentially returning large result sets. The tool is read-only (good security posture), but overall definition completeness is mediocre.

• Add a formal output schema to the tool description: 'Returns a list of yield pool objects with fields: chain (string), pool (string), project (string), tvlUsd (float), apy (float), apyMean30d (float), predictions (object).'
• Replace 'Optional filter for blockchain' with 'Optional blockchain filter (case-insensitive). Valid examples: Ethereum, Solana, Polygon, Arbitrum, Optimism, Base. Leave empty to include all chains.'
• Replace 'Optional filter for project name' with 'Optional project filter (case-insensitive). Valid examples: lido, aave-v3, curve, yearn. Leave empty to include all projects.'
• Add limit and offset parameters: 'limit (int, default 20, max 100): maximum pools to return' and 'offset (int, default 0): pagination offset. Use limit + offset or implement a next_cursor field in the response.'
• Cap results at 20 - 50 pools by default and update the description: 'Returns up to 20 yield pools. Use limit and offset parameters to retrieve more.'
• Enhance error handling to return specific, actionable messages: 'API unavailable (yields.llama.fi returned status {status}). Retry in 30 seconds or check service status.' Instead of re-raising, return a structured error with retry guidance.
• Document the response format with a return type annotation or docstring example showing the exact field structure each pool object will have.
• Consider returning IDs or unique identifiers (pool_id, project_id) if the agent will need to chain calls (e.g., get detailed info on a specific pool).

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry