Loading ToolBench
URL or long-form text to social carousel CLI and MCP client.
ClawVisual has 4 well-structured tools with complete input schemas and mostly solid descriptions. All tools use appropriate verb-noun naming patterns (convert, job_status, revise, regenerate_cover). Schemas are comprehensive with proper JSON Schema typing, enums, and constraints. However, descriptions lack depth on WHEN to use each tool vs. alternatives, error recovery paths are not documented, and output schemas are not visible in the source code. Tool composition is reasonable but interdependencies (e.g., convert returns job_id needed by job_status) are not explicitly documented in descriptions. Parameter descriptions are present and reasonably detailed (average ~50-80 chars), meeting baseline expectations. No security issues detected (no secrets in params). The 'convert' tool is complex but well-parameterized; 'revise' and 'regenerate_cover' have some overlapping intent that could cause LLM confusion.
Start long-text to short-carousel conversion and return job id.
Get latest status and result of a conversion/revision job.
Regenerate cover either from an existing job or by direct prompt.
Create a revision job for copy rewrite or image regeneration.
Output schemas not documented in source code. Tool descriptions state what is returned (e.g., 'return job id') but there is no visible JSON Schema or structured documentation of response fields. LLMs cannot plan downstream tool calls without knowing what fields to expect.
Tool descriptions lack WHEN guidance. Descriptions state WHAT each tool does but do not explain when to call it instead of a similar tool. The distinction between 'revise' (modify copy/images) and 'regenerate_cover' (regenerate cover image specifically) is unclear and may cause LLM to conflate them.
Error handling and recovery guidance not visible. No documented error codes, categorization (retryable vs. fatal), or recovery paths (e.g., 'If conversion fails, retry with fewer slides'). Agents will not know how to handle failure states.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 12 | 2025-06-18+ | v2 |
| 2026-03-09 | D | 51 | - | v1 |
Tool chain dependencies not documented. 'convert' returns job_id, which is required by 'job_status' and 'revise', but descriptions do not state this dependency or guide the LLM on multi-step workflows. Agents may not understand the natural call sequence.
'revise' tool has mutually exclusive or dependent parameters that are not clearly documented. Parameters like 'intent' enum and corresponding 'editable_fields' or 'mode' have implicit dependencies (e.g., certain intents require certain fields) that should be stated explicitly in descriptions to prevent invalid combinations.
Parameter 'target_slides' in 'convert' is marked deprecated but still present. Deprecation path is unclear, should LLM prefer 'max_slides'? Should 'target_slides' be removed entirely? Presence of deprecated params increases confusion and token cost.