ClauseAI has 3 tools with clear verb-noun naming (list_, get_, generate_). Descriptions are present and reasonably detailed (100-150 chars each). Input schemas are visible with type declarations. However, parameter descriptions are sparse or missing entirely, 'slug' and 'answers' lack guidance on format/constraints. Output schemas are not documented. Error handling is absent from tool definitions. The generate_document tool accepts 'answers' as a free-form object with no guidance on required vs optional fields or validation rules, forcing LLMs to guess. No tool annotations (readOnlyHint/destructiveHint) despite clear risk levels.

• Add detailed parameter descriptions to 'slug' (e.g., 'Template identifier returned by list_templates. Format: lowercase alphanumeric with hyphens, 2-50 chars.') and 'answers' (e.g., 'Object mapping field keys (from get_template_fields) to string values. Missing keys remain as placeholders. Invalid keys are ignored.').
• Document output schemas for all three tools. For list_templates, specify the structure of each template object (id, name, description, fields). For generate_document, return {file_path, file_size, format, generation_id} or similar.
• Add tool annotations: mark generate_document with destructiveHint=true (creates a record), list_templates and get_template_fields with readOnlyHint=true.
• Include error guidance in descriptions: 'If slug is invalid, call list_templates() to discover valid template identifiers.' For generate_document, clarify what happens with incomplete answers (e.g., 'Missing fields remain as [FIELD_NAME] placeholders in the output').
• Specify constraints on 'format' parameter: use an enum (pdf|odt|markdown) rather than free-form string to prevent hallucinated formats.