Loading ToolBench
Kubernetes cluster monitoring and diagnosis agent with Google Chat integration. Watches for failing workloads, runs automated diagnoses via Hermes, and delivers reports.
Leandro exposes 2 tools with minimal definition quality. Neither tool has a visible input schema in the source code, only bash script file paths are referenced. Descriptions exist but are vague and lack actionable detail for LLM selection. No parameter documentation, no output schema, no error handling guidance. Tool names lack action verbs (cluster-snapshot, session_search). This is a shell-script-based system where tool definitions appear to be inferred rather than explicitly registered with proper MCP schemas.
Compact read-only cluster state snapshot for cron prompt injection. Collects Kubernetes cluster facts: NotReady nodes, non-Running pods (top 15 by restarts), warning events from last hour, and unbound PVCs. Output is raw facts only for agent interpretation.
Search and analyze session history over a time period. Used to list incidents from the week, identify recurring patterns and trends in diagnoses and chat sessions.
No input schemas visible for either tool. Tool definitions appear to be inferred from bash script file paths rather than explicitly registered with JSON Schema.
Tool names lack action verbs. 'cluster-snapshot' and 'session_search' do not start with verbs (get_, list_, search_, etc.). LLMs cannot infer intent from names alone. Should be 'get_cluster_snapshot' or 'search_sessions'.
Descriptions lack actionable detail. 'cluster-snapshot' says 'Compact read-only cluster state snapshot for cron prompt injection' but does not explain WHEN to call it, WHAT parameters it accepts, or WHAT the output structure is. 'session_search' description is similarly vague about parameters and return format.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 32 | 2026-07-28+ | v2 |
No parameter documentation. Neither tool shows input parameters, their types, constraints, or descriptions. LLMs cannot determine what to pass or how to construct valid calls.
No output schema or return type documentation. LLMs cannot plan downstream tool calls or extract required fields (e.g., does session_search return incident IDs, timestamps, summaries?). Bash script implementations suggest raw text output, not structured JSON.