Scoring was not performed

• Extract parameter schemas from tool descriptions and define them formally in JSON Schema with types (string, number, array, object), constraints (min, max, pattern, enum), and required fields. Make each parameter independently documented.
• Rewrite tool descriptions to be concise summaries (50-150 chars) answering: What does it do? When should the LLM call it? Replace 'Choose the right approach...' procedural text with a short, actionable statement.
• Document all output schemas explicitly. For discover_data, specify what fields are returned: asset_id, name, type, provider, metadata, ownership_info, etc. Include field types and whether results are paginated.
• Add error handling guidance to descriptions. E.g., 'Returns empty results if no matches. Use broader filters or different query terms to discover more assets.'
• Remove embedded examples (literal values like 'customer', 'kafka', 'postgres') from descriptions. Replace with formal enum constraints in the parameter schema.
• Define pagination behavior clearly: does discover_data support offset/limit? What is the default and maximum page size? Return total count or next_cursor?
• For find_ownership, clarify whether asset_id is required or optional. Can all three (asset_id, user_id, team_id) be passed simultaneously, or are they mutually exclusive? State explicitly in parameter descriptions.
• Add parameter examples as schema constraints (enum, pattern) rather than narrative examples. A 'types' enum value list is machine-readable and prevents LLM hallucination.