Loading ToolBench
Multi-agent orchestration MCP server with file extraction, web crawling, web search, plugin factory, and job scheduling capabilities
Derrick has 10 tools with schemas and descriptions present, but significant quality gaps limit production readiness. Tool names follow verb_noun convention (files_extract, agents_spawn, jobs_create), which is good. However, descriptions are often vague or incomplete. For example, 'files_extract' says 'extract (text/markdown) or convert (csv↔xlsx, pdf/docx/html→markdown)' but doesn't explain WHEN to use it or what it returns. Parameter descriptions exist but lack specificity: 'timeout_seconds' has a range (1...180) but no guidance on when to increase it. No output schemas are documented anywhere in the source, LLMs cannot predict what these tools return. Error handling is minimal; no recovery guidance or categorization. Security concerns: agents_spawn, agents_send, jobs_create, and plugin_factory_build are WRITE operations but lack permission gates or audit trail documentation. The orchestration tools (agents_*) expose agent IDs as strings with no validation or sanitization guidance. Overall, definitions are skeletal, present but not LLM-optimized.
Delegate a task to a specific agent profile (developer, researcher, or general)
Cancel an agent
Complete a task and return result to parent agent
List agents in the orchestration hierarchy
Send a message to another agent (parent or child)
Spawn a worker agent with a goal and task
extract (text/markdown) or convert (csv↔xlsx, pdf/docx/html→markdown)
Create a durable job or schedule order for delayed/scheduled execution
No output schemas documented. LLMs cannot predict return types, forcing them to guess what fields are available for downstream tool calls. This breaks tool chaining and forces discovery detours.
Descriptions lack WHEN/WHY guidance. 'Spawn a worker agent with a goal and task' doesn't explain when to use agents_spawn vs agent_profile_delegate, or what happens to the spawned agent's output. LLMs cannot disambiguate.
Write operations (agents_spawn, agents_send, agents_cancel, jobs_create, jobs_schedule_create, plugin_factory_build) lack permission gates and audit trail documentation. No indication of what permissions are required or how to trace who called what.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | 2026-07-28+ | v2 |
Create a recurring or one-time schedule for job execution
Build an Agent Plugin from a user goal
Parameter descriptions lack format/constraint details. 'agent_id' is described as 'Optional stable worker id (slug)' but no regex, length limit, or character restrictions are stated. LLMs cannot validate inputs.
No error handling guidance. Tools do not document what errors are retryable, user-fixable, or fatal. No recovery suggestions (e.g., 'If agent not found, call agents_list() first').