Loading ToolBench
Context-aware clickable choices that resolve material ambiguity before an AI assistant acts.
Single tool 'ask_choice_question' has a well-structured schema with comprehensive parameter definitions, clear descriptions, and proper constraints (min/max lengths, array bounds). Naming is verb-noun compliant ('ask_choice_question'). Description is detailed (180+ chars) and explains the purpose and constraints. Input schema is fully visible with typed fields, descriptions, and validation rules. However, output schema is not documented, the tool returns user-selected choice text but the response structure is not formally specified. Error handling guidance is minimal. Overall solid for a single-purpose tool but lacks output documentation and recovery guidance.
Render one context-aware decision question with two to five clickable options. Each value must be a concise, actionable user instruction. Recommend at most one option and allow custom input only when useful.
Output schema not documented. Tool returns user choice but response structure (fields, types, error cases) is not formally specified. LLMs cannot plan downstream actions without knowing what fields to expect.
No error handling guidance. Tool description does not explain failure modes (e.g., invalid custom input, timeout, host unavailable) or recovery steps. LLMs lack actionable error recovery paths.
Missing idempotency guidance. Tool modifies agent state (sends user message) but does not document whether repeated calls with identical input are safe or produce duplicate side effects.
No permission or scope declaration. Tool sends messages on behalf of the user but does not declare required permissions (e.g., 'send:message', 'interact:ui'). Audit trail requirements unclear.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 58 | 2025-06-18+ | v2 |