ModelForge exposes 22 tools with reasonable naming (verb-first: list_, get_, build_, export_, etc.) and descriptions present for all tools. However, critical gaps undermine quality: (1) Input schemas are visible but lack detailed parameter descriptions, most params have only type and brief text, missing constraints, ranges, and format guidance. (2) Output schemas are NOT documented anywhere in the source; LLMs cannot plan downstream calls or extract structured data. (3) No error handling guidance, tools return success/failure but do not guide recovery (e.g., 'If validation fails, call validate_spec first'). (4) Parameter relationships undocumented (e.g., build_model's optional output_dir and model_type override lack dependency hints). (5) Tool descriptions average ~80 chars, which is adequate but lack WHEN-to-use context and prerequisites. Example: 'build_model' says 'YAML spec → live-formula Excel workbook + SQLite linkage graph' but does not explain when to call it vs certify(), or what happens if the spec is invalid. Per-tool analysis: list_templates, spec_guide, get_spec_schema, validate_spec, build_model, certify, qc_workbook, list_sources, lineage_walk, ingest_dataroom, export_pptx, export_deck, screen_deals, compute_tax, export_docx, data_providers_status, quote, history, fundamentals, search_filings, entity_lookup, search_securities all follow similar pattern, good naming, minimal descriptions, no output schema docs, no error recovery guidance.

• Document output schemas for all 22 tools. For each, specify: field names, types (string|number|object|array), required vs optional, and examples. E.g., list_templates should return {templates: [{name: string, description: string, model_type: string}], total: number}.
• Add parameter constraints to descriptions. For 'model_type', list valid enum values (e.g., 'one of: dcf, lbo, merger, project_finance'). For 'output_dir', specify format ('absolute or relative path; defaults to ./output'). For 'limit', add range ('1 - 100, default 20').
• Add error recovery hints to tool descriptions. E.g., validate_spec: 'Returns validation errors with line numbers. If errors occur, review spec_guide() for the model_type and fix the YAML, then retry.' build_model: 'If build fails, call validate_spec() first to identify issues.'
• Clarify tool selection logic in descriptions. For overlapping tools (build_model vs certify, export_pptx vs export_docx), add a 'Use this tool when...' sentence. E.g., 'Use export_pptx for executive summaries; use export_docx for detailed credit memos.'
• Add rate-limit and timeout guidance to data provider tools. E.g., quote: 'Returns latest price (updated every 15 min). Rate limit: 100 calls/min. Timeout: 5s. If unavailable, returns {error: "provider_down", fallback_source: "..."}'.