This is a metrics sink plugin with 2 tools. Both tools have descriptions, but they are domain-specific and lack LLM-optimization guidance. The 'emit' tool description (67 chars) explains the accumulation behavior but does not state WHEN to call it or what prerequisites exist. The 'render' tool description (145 chars) is better but still lacks clarity on error cases. Neither tool has visible input parameter descriptions in the schema, the 'metric' parameter for 'emit' has a type hint ('MetricPoint') but no description explaining what fields it expects or how to construct it. The 'render' tool accepts no parameters, which is correct, but the overall schema documentation is minimal. No output schemas are documented. This is a plugin designed for operator/gateway integration, not LLM agent interaction, which explains the gap between tool definitions and agentic patterns.

• Add a description to the 'metric' parameter: 'A MetricPoint event containing the metric name, kind (counter/gauge/histogram), labels, and value(s). On kind mismatch with existing series, the emit call fails and the new sample is dropped.'
• Document the return value for 'emit': 'Returns {success: true} on accumulation, or {success: false, error: string} if the metric kind conflicts with an existing series of the same name.'
• Document the return value for 'render': 'Returns a string in Prometheus text-exposition v0.0.4 format (lines of metric_name{labels} value). Safe to call concurrently; returns a snapshot of the current registry state.'
• Add WHEN guidance to 'emit': 'Call this tool each time a metric event occurs (counter increment, gauge update, histogram observation). Typically invoked by the gateway's metrics-rs bridge, not directly by agents.'
• Add error recovery guidance to 'emit': 'If you receive a kind mismatch error, verify the metric name and type match previous emissions. Switching from counter to gauge on the same name is not allowed; use a different metric name if the type must change.'