Loading ToolBench

Reading the current corpus…

OpenHuskyAgent MCP server · D · ToolBench

OpenHuskyAgent

A multi-agent orchestration system with tool delegation, subagent management, and MCP client integration for LLM-powered task execution

2026-07-28+1 tools

50/100

Definition Quality64

Protocol Readiness35

Supportability35

Spec posture

Inferred effective spec 2026-07-28+·current baseline 2026-07-28

Static source inference · medium confidence

Current-spec patterns detected

Summary

Single tool 'delegate_task' has a detailed input schema with 6 parameters, all typed and described. Description is comprehensive (280+ chars) and explains single/parallel execution, timeouts, and toolset restrictions. However, output schema is completely undocumented, no return type specification visible. Parameter descriptions are adequate but lack examples of valid toolset values and error recovery guidance. The tool combines multiple concerns (single task, parallel tasks, context passing) which could be split for clarity. No evidence of error handling patterns or recovery guidance in the schema.

Tools (1)

delegate_taskwritesource verified64/100

Delegates task(s) to focused subagents with isolated context and restricted tools. Single task: pass 'goal'. Parallel tasks: pass 'tasks' array (up to 10 concurrent). Use timeout_seconds when the task is expected to need more or less time; default is 600 seconds. Each subagent runs in its own session and returns a structured result.

Issues (5)

CRITICALresponse-shaper ↗

Output schema completely undocumented. No return type, field structure, or success/failure format specified. LLMs cannot plan downstream operations or extract results.

delegate_task

HIGHrecovery-guide ↗

No error handling guidance. Tool description lacks recovery instructions for timeout, invalid toolset, or subagent failure scenarios. LLMs cannot self-correct on errors.

delegate_task

HIGHconstrained-input ↗

Parameter 'required_toolsets' lacks enum constraint. Description lists available options as prose ('CORE, TERMINAL, SEARCH...') instead of formal enum. LLMs may hallucinate invalid toolset names.

delegate_task

MEDIUMtool ↗

Tool combines single-task and parallel-task modes in one definition. Consider splitting into 'delegate_single_task' and 'delegate_parallel_tasks' for clarity and reduced cognitive load on LLM selection.

delegate_task

Recommendations

Document the complete output schema: define return type (object/array), all fields (e.g., task_id, status, result, error), and their types. Example: {"type": "object", "properties": {"task_id": {"type": "string"}, "status": {"type": "string", "enum": ["success", "failed", "timeout"]}, "result": {"type": "object"}, "error": {"type": "string"}}}
Add error handling guidance to description: 'On timeout, retry with increased timeout_seconds. On invalid toolset, call list_available_toolsets(). On subagent failure, check subagent logs via get_subagent_status().'
Convert 'required_toolsets' to a formal enum in the schema: {"type": "array", "items": {"type": "string", "enum": ["CORE", "TERMINAL", "SEARCH", "WEB", "BROWSER", "MEMORY", "KNOWLEDGE", "EXECUTE", "DELEGATE", "SKILLS", "MCP", "BUSINESS", "VISION"]}}
Explicitly state in parameter descriptions: 'goal and tasks are mutually exclusive, pass exactly one. goal for single task, tasks for parallel execution (up to 10 concurrent).'
Add per-item success/failure reporting for parallel tasks: return array with per-task status, result, and error message so LLM knows which tasks succeeded and which failed.

Score history

Overall score trend

First recorded score · v2 rubric

50/100

Scored	Grade	Overall	Spec posture	Rubric
2026-09-23	D	50	2026-07-28+	v2