Two tools with complete input schemas and reasonable descriptions, but significant gaps in error guidance, output documentation, and parameter constraints. Tool names follow verb_noun convention (generate_*, edit_*). Descriptions are adequate (80-90 chars) but lack context on when to use each tool or prerequisites. Parameters have types and basic descriptions, but lack enums, ranges, or format constraints. No documented output schema. Error handling returns generic exceptions without recovery guidance. Missing tool annotations (destructiveHint, idempotentHint). No pagination or result limits documented.

• Document output schema for both tools: {file: string, asset_descriptor: {asset_id, variants, attribution, extra}}. Include this in tool definition so LLM knows what fields to expect.
• Add enums to model, size, quality parameters. E.g. model: enum [dall-e-3, dall-e-2], size: enum [1024x1024, 1024x1792, 1792x1024, 512x512], quality: enum [standard, hd].
• Enhance error messages with recovery guidance. E.g. 'OPENAI_API_KEY not set. Configure via environment variable or server config.' or 'OpenAI API rate limit exceeded. Retry in 60 seconds.' or 'Invalid prompt: too long (max 4000 chars).'
• Add destructiveHint and idempotentHint annotations to tool definitions. Both tools write files and call external APIs, mark as destructive. Idempotent if same prompt+model+size always produces same image (likely not, OpenAI API is non-deterministic).
• Validate OPENAI_API_KEY at server initialization, not per-request. Return clear error to LLM if missing: 'OpenAI API key not configured. Contact administrator.' Prevents repeated failures.