Two tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. play_rps has a well-formed input schema with enums and defaults, but get_stats lacks any parameter descriptions. Neither tool documents return types or error handling. Descriptions are present but generic (194 chars baseline; these are 80-90 chars). No tool annotations, no error recovery guidance, no output schema documentation. Naming follows verb_noun convention (play_rps, get_stats) which is good, but descriptions don't explain WHEN to use each tool or what the LLM should expect in responses.

• Expand get_stats description to 100-150 chars: 'Retrieve cumulative game statistics including wins, losses, and ties. Call this after playing to see updated stats, or before playing to establish a baseline.'
• Add parameter description for 'difficulty' in play_rps: 'AI difficulty level. easy=random moves, medium=frequency analysis, hard=multi-tier strategy with win-stay/lose-shift detection, veryhard=reserved for future use.'
• Document output schema for play_rps: {result: 'win'|'loss'|'tie', player_choice: string, computer_choice: string, winner: string, stats: {wins: int, losses: int, ties: int}}
• Add error handling: return {error: 'Invalid choice. Must be one of: rock, paper, scissors', suggestion: 'Call get_stats() first to verify game state'} on invalid input.
• Add tool annotations to schema: readOnlyHint=true for get_stats; idempotentHint=true for play_rps (same input always produces same outcome given same game history).