Loading ToolBench
xUdav: local LM Studio workspace, CLI, and MCP-oriented dev layout.
Server has 13 tools with basic schemas and descriptions, but significant quality gaps. Most tool descriptions are terse (10-50 chars), well below the 50-200 char LLM-optimized baseline. Parameter descriptions are sparse or missing entirely. Many tools lack clear error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Schemas are present but minimal, most parameters lack detailed constraints, enums, or format specifications. The finalize-task tool has complex interdependent parameters with no documented relationships. Overall structure is functional but falls short of production-grade agent tool standards.
Add project to registry.
Create baseline directories and registry.
Build session brief for a new chat (data/context/briefs/…).
Finalize task status against a completion contract.
Build file index for active or chosen project.
List registered projects.
Append decision to decision log.
Descriptions are uniformly terse (10-50 chars), far below the 50-200 char LLM-optimized baseline. Examples: 'Create baseline directories and registry' (43 chars), 'Show system status' (18 chars). LLMs cannot determine when/why to select tools without richer context.
finalize-task has 13 interdependent parameters (step-index, max-steps, action-fingerprint, repeated-fingerprint-count, etc.) with no documented relationships or validation rules. LLMs cannot reason about which combinations are valid.
No tool annotations present (readOnlyHint, destructiveHint, idempotentHint). Tools marked WRITE risk (bootstrap, sync-mcp, add-project, switch-project, index-project, rebuild-context, log-decision, brief, prepare-chat) should declare destructiveHint=true to guide agent retry logic.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 51 | 2026-07-28+ | v2 |
Switch project (optional), index, sync-mcp and build brief in one command.
Rebuild context for active or chosen project.
Search in active or chosen project.
Show system status.
Switch active project.
Write config/mcp/mcp.json for the active project (filesystem + shell scope).
Parameters lack enums and format constraints. 'mode' in search-project accepts 'file, text, todo, entrypoints, or defs' but is defined as free-form string. 'format' in brief accepts 'short or full' but no enum. LLMs will hallucinate invalid values.
No error handling guidance in any tool description. Tools do not explain what to do on failure (retry, ask user, abort). Agents lack recovery paths when tools fail.