Loading ToolBench
OpenAI Function Calling bridge for Agenium agent:// protocol
This server exposes 6 tools for bridging OpenAI function calls to Agenium agents. While tool names follow verb_noun convention (discover_tools, execute, register, resolve, skill_to_tool, agent_card_to_tools), descriptions are minimal and parameter documentation is severely lacking. Most tools lack descriptions of what they return, and parameter descriptions are generic or missing entirely. No error handling guidance, no output schema documentation, and no validation constraints visible. The schema for 'execute' accepts 'Any' type, which violates JSON Schema best practices. Overall, this reads as a proof-of-concept rather than production-grade tooling.
Convert all skills in an AgentCard to OpenAI tool definitions. If the card has no skills, a single generic send_message tool is created
Resolve agent_uris, fetch their agent cards, and return OpenAI tool definitions
Execute OpenAI tool call(s) against their respective agents
Register this bridge's agent on the Agenium DNS system
Resolve agent:// URIs via the Agenium DNS HTTP API
Convert a single AgentSkill to an OpenAI tool dict
execute tool accepts 'Any' type for tool_calls parameter, violating JSON Schema constraints. LLMs cannot validate input shape before calling.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields (e.g., agent_uri, endpoint, skill_name).
Parameter descriptions are generic or missing. 'agent_uris' lacks format guidance; 'tool_calls' lacks structure; 'skill' parameter type is 'AgentSkill' with no description of required fields.
No error handling guidance. Tools do not document what errors are retryable, what user-fixable errors look like, or how to recover from DNS resolution failures or agent timeouts.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | <=2025-11-25 | v2 |
Tool descriptions do not state whether operations are idempotent or have side effects. 'register' and 'execute' modify state but lack warnings about retry safety.