Loading ToolBench
A server that provides tools for generating and saving images using DALL-E 3
Two tools with complete input schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. generate-image has well-defined enums and constraints; save-images lacks descriptions for individual parameters. Neither tool documents output structure for LLM planning. Error handling is present but generic. No tool annotations (readOnlyHint/destructiveHint). Naming follows verb_noun pattern correctly. Average parameter description coverage ~60%.
Generate images using OpenAI's DALL-E 3 model based on text prompts
Save generated images from URLs to local filesystem
Output schemas not documented. LLMs cannot plan downstream operations or extract required fields. generate-image returns {urls, revised_prompts} but this is only visible in code, not in tool definition. save-images returns {saved_paths, message}, also undocumented.
save-images parameters lack individual descriptions. 'urls' array, 'prompt', 'revisedPrompts', 'outputDir' are defined in schema but have no per-parameter guidance text. LLMs cannot infer when revisedPrompts is required vs optional or what format outputDir expects.
No tool annotations (destructiveHint, readOnlyHint, idempotentHint). save-images writes to filesystem, should be marked destructive. generate-image is read-only but unmarked. Agents cannot reason about side effects or retry safety.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 48 | <=2025-11-25 | v2 |
Error messages are generic ('Image generation failed: {error.message}', 'Failed to save images: {error.message}'). No recovery guidance. If OpenAI API rate-limits or auth fails, LLM receives raw error with no hint to retry, check credentials, or call alternative tool.
generate-image description does not mention that n=1 is typical for DALL-E 3 (model limitation), or that revised_prompts may differ from input. save-images description does not clarify that outputDir defaults to './images' or that metadata JSON files are auto-created alongside images.