Loading ToolBench
A Rust-based agent framework with memory management, code indexing, execution pipeline, and daemon services. Includes MCP tool definitions for memory operations, search, context planning, code execution, and observability.
HZR has 14 tools with complete JSON schemas and basic descriptions, but falls short of production quality. Descriptions are minimal (10-50 chars), lacking WHEN/WHY context and recovery guidance. Parameter descriptions are absent entirely, LLMs cannot infer what 'filter' or 'query' mean. No output schemas documented. Error handling is not visible in tool definitions. Tool names are verb-prefixed but lack clarity: 'hzr_memory_recall' vs 'hzr_memory_get' are confusingly similar. Destructive tools (hzr_memory_forget, hzr_memory_prune) have no confirmation or dry-run pattern. Security-sensitive tools like hzr_exec and hzr_write lack permission gates or audit trail documentation.
Encode or decode content using HZR codec
Plan and retrieve context candidates for code understanding
Run diagnostic checks on the HZR system and environment
Execute commands in the workspace with controlled environment
Delete a memory record by ID
Get a specific memory record by ID
Prune memory records based on criteria
Parameter descriptions completely absent. Tools like hzr_search have 'query' and 'paths' with no explanation of format, constraints, or examples. LLMs cannot infer whether 'query' is a regex, glob, or literal string.
Tool descriptions are too brief (10-25 chars). 'Recall memory records from the memory system' and 'Execute commands in the workspace with controlled environment' lack WHEN to use, prerequisites, and what the tool returns. Descriptions should be 50-200 chars with actionable context.
Destructive tools (hzr_memory_forget, hzr_memory_prune, hzr_write, hzr_exec) lack confirmation or dry-run patterns. Agents can delete records or execute arbitrary commands without a safety gate. No error recovery guidance visible.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
Recall memory records from the memory system
Store a new memory record
Update an existing memory record
Query observability and diagnostic information about the system
Read file content from the workspace
Search for files and content matching query patterns
Write content to a file in the workspace
Output schemas not documented. Tools return results but LLMs have no specification of response fields, types, or structure. This forces LLMs to guess what data is available for downstream tool calls.
Similar tool names create ambiguity: hzr_memory_recall vs hzr_memory_get both retrieve memory, but distinction is unclear. hzr_memory_recall has no parameters; hzr_memory_get requires an ID. Descriptions do not explain when to use each.