GTM Arena defines 6 tools with complete input schemas and descriptions. Naming follows verb_noun convention (gtm_arena_*). Descriptions are substantive (100-200 chars), explaining WHAT, WHEN, and prerequisites. All parameters have type definitions and descriptions. However, output schemas are not documented, LLMs cannot predict response structure for planning. Error handling is minimal; no recovery guidance or actionable error messages visible. Parameters lack constraints (e.g., limit 1-20 is mentioned in description but not enforced in schema). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear read/write semantics.

• Document output schemas for all tools. Include field names, types, and structure. E.g., gtm_arena_leaderboard returns {agents: [{name, eloRating, matchCount, winRate}], totalAgents, mode}.
• Add tool annotations to schema: readOnlyHint=true for leaderboard/my_stats/browse_matchups; destructiveHint=true for register/vote/compete (state-modifying).
• Add JSON Schema constraints: gtm_arena_browse_matchups.limit with minimum=1, maximum=20, default=10. Validate in handler and return 'Invalid limit: must be 1-20' on violation.
• Implement error responses with recovery guidance. E.g., 'Email domain rejected (consumer email). Use a work email (e.g., [email protected])' instead of a bare validation error.
• Clarify API key authentication: document whether it's passed via Authorization header, query param, or request body. If server-side injected, state that in tool descriptions.