Knowledge Autoload & Health Inspection Improvements

Health Inspection Improvements

Health inspection now has a single, consistent entry point: whether you start it with the /health command or through inspection mode, it runs the same flow.

Inspection reports are reorganized by knowledge category for a clearer structure, so you can quickly find the checks and conclusions you care about.

During an inspection, the agent can also draw on past inspection reports, producing more grounded conclusions and making it easier to compare system state and trends across different points in time.

Full Release Checklist

feat

• Added an autoload option for knowledge: only summaries are injected by default, with full content loaded automatically in applicable scenarios when enabled.
• Improved health inspection: unified the /health command and inspection mode, added knowledge categorization with a refined display structure, and made past inspection reports available for reference.
• Improved best practices and diagnostics for Grafana, Loki, Tempo, Prometheus, Elasticsearch, ClickHouse, and Zabbix connectors.
• Document authoring now opens the matching preview panel by content type and streams content in during creation.
• Unified tool call and visualization reference identifiers for more stable, consistent reference navigation.
• Multiple UI and interaction refinements: task detail typography, sidebar overlay, input box shadow and focus states, inline code wrapping, and task interruption status display.

bugfix

• Fixed a statistics error caused by token usage when the total was empty.