Loading ToolBench
Governance-native Agents-as-a-Service runtime with spectrum refraction capabilities, supporting multiple MCP servers and tools for calendar, email, image generation, and more.
Critical failures across all dimensions. Only 3 of 9 tools have visible input schemas (calendar_plan, email_ops, image_generate). The 6 scraping tools are defined only in package.json with names and generic descriptions, no input schemas, no parameter documentation. Tool descriptions are present but generic (10-50 chars for scrapers). No output schemas documented. No error handling guidance. Naming is inconsistent: 'calendar_plan' and 'email_ops' use underscores but lack verb clarity; 'scrape_*' tools lack parameter documentation entirely. The server appears to be a collection of partially-implemented adapters with significant gaps in schema completeness and parameter validation.
File-backed calendar event management tool supporting availability lookup, event creation/update, proposals, date range queries, and event deletion. Integrates with Google Calendar and Outlook when OAuth adapters are connected.
File-backed email triage and draft tool supporting thread summarization, draft reply composition, email sending, inbox listing, and thread management (mark read, archive). Integrates with Gmail and Outlook when OAuth adapters are connected.
Generates images via configured provider (OpenAI, OpenRouter, or Gemini) based on text prompts. Saves resulting images to workspace and returns saved path. Provider-agnostic dispatch with model-capability matrix routing.
Get educational arXiv papers (open access)
Access US government educational resources (public domain)
6 of 9 tools have NO input schemas. Scraping tools (scrape_mit_ocw, scrape_wikipedia_educational, scrape_arxiv_papers, scrape_government_educational, scrape_khan_academy_cc, validate_license_compliance) are defined only in package.json with names and descriptions, no parameter documentation, no type information, no constraints. LLMs cannot determine what parameters to pass or what values are valid.
Scraping tool descriptions are generic and under 20 characters (e.g., 'Scrape MIT OpenCourseWare content (Creative Commons licensed)'). No guidance on when to use each tool, what parameters they accept, or what they return. Descriptions must explain WHAT, WHEN, and WHAT IT RETURNS.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 40 | <=2025-11-25 | v2 |
Access Khan Academy Creative Commons content
Scrape MIT OpenCourseWare content (Creative Commons licensed)
Extract educational Wikipedia articles (CC-BY-SA licensed)
Validate that scraped content is license-compliant
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. calendar_plan, email_ops, and image_generate have input schemas but no documented return types or field structures.
calendar_plan and email_ops accept 'action' as a free-form string enum but do not validate against a strict set of allowed values. LLMs may hallucinate invalid actions. Should declare action as a constrained enum: ['availability_lookup', 'create_or_update_event', 'propose', 'list_range', 'delete_event'] for calendar_plan.
No error handling guidance. Tools do not document what errors can occur, whether they are retryable, or what the LLM should do next. E.g., calendar_plan may fail if calendarId is invalid or event conflicts exist, but no recovery guidance is provided.