Loading ToolBench
MCP Server for reordering documents using U-shaped attention curve to eliminate lost-in-the-middle effect in LLM context windows
Single tool with minimal definition quality. Tool name 'reorder_context' lacks action verb prefix (should be 'reorder_documents' or similar). Description is adequate (89 chars, within 10-1024 range) but lacks WHEN/WHY guidance. Input schema present but severely incomplete: 'documents' array items are typed as generic 'object' with no property descriptions, no constraints on document structure, no required fields within items. No output schema documented. No parameter descriptions. No error handling guidance. No examples of expected input/output format.
Arrange documents in U-shaped attention curve to eliminate lost-in-the-middle effect.
Tool name lacks action verb. 'reorder_context' is ambiguous; should be 'reorder_documents' or 'reorder_by_attention_curve' to clearly signal the action.
Input schema 'documents' array items are untyped objects with no property descriptions. LLM cannot infer required fields (id? score? content?), data types, or constraints. Schema must define each property with type and description.
No output schema documented. LLM cannot predict return structure, field names, or types. Must document: returns array of reordered documents with same structure as input, or specify transformed fields.
Parameter 'documents' lacks description. LLM does not know: required fields per document, expected score range, how missing scores are handled, max array length, or whether documents are modified in-place.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 38 | 2026-07-28+ | v2 |
No error handling guidance. Code silently returns original list if len(documents) <= 2; no error for missing 'score' key (defaults to 0.0). LLM needs explicit error messages and recovery paths.