Loading ToolBench
Generates design artifacts (pitch decks, landing pages, dashboards, emails, social posts, animations) via Gemini AI, with PDF rendering via wkhtmltopdf and PowerPoint generation via python-pptx
Server has 9 tools with schemas and descriptions present, but quality is significantly below production baseline. Tool names lack clear verb-noun structure (e.g., 'generate_pitch_deck' vs 'create_pitch_deck'). Descriptions are present but generic (avg ~80 chars, below 194-char baseline). Parameter descriptions are minimal or absent for key fields like 'design_system' (only 'Design system ID' with no guidance on the 150 available options). No output schemas documented, tools return file paths as strings with no structure. Error handling is minimal (generic try-catch in server.py). No tool annotations (readOnlyHint, destructiveHint). All 7 generation tools are WRITE operations but lack confirmation/dry-run patterns. Catalog tools (list_design_systems, list_skills) are READ-only but lack pagination or result limits.
Generate an analytics/admin dashboard design as HTML. Returns the HTML file path for sending.
Generate a branded email template as HTML. Returns the HTML file path for sending.
Generate an HTML animation/motion graphic (HyperFrames format) with GSAP. Returns HTML file path. Can be opened in browser for animated preview.
Generate a landing page design as HTML. Returns the HTML file path for sending.
Generate a pitch deck HTML for a product/project. Returns the HTML file path for sending.
Generate a PowerPoint (.pptx) pitch deck file. Returns the PPTX file path for sending.
No output schemas documented. All generation tools return file paths as plain strings with no structure (e.g., 'Returns the HTML file path for sending'). LLMs cannot infer what fields to expect or how to chain results.
Parameter 'design_system' lacks actionable description. Docs say '150 available' but do not list them or explain how to discover them. LLMs cannot pick valid values without calling list_design_systems first, forcing extra round-trips.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | 2026-07-28+ | v2 |
Generate a social media post design as HTML. Returns the HTML file path for sending.
List all available design systems for generation
List all available generation skills
No tool annotations (readOnlyHint, destructiveHint, idempotentHint). All 7 generation tools are WRITE operations that create files, but LLMs have no signal that these are destructive or non-idempotent. Agents may retry blindly, creating duplicates.
Catalog tools (list_design_systems, list_skills) have no pagination, limit, or result count. If 150 design systems exist, returning all at once wastes tokens and risks context window exhaustion.
Error handling is generic (try-catch returns 'Error: {type}: {message}'). No recovery guidance, no categorization of retryable vs fatal errors, no actionable next steps for LLMs.