A cloud-edge integrated observability and AIOps platform with multi-language APM support, alert management, and AI-driven incident response
Critical gaps across all dimensions. Tool definitions are inferred from test files and scattered source locations rather than centralized, explicit registration. No input schemas visible for 12 of 14 tools. Descriptions are present but generic (10-50 chars typical). No parameter type definitions, no output schemas documented, no error handling guidance. Tools like 'host_bash' and 'ExecuteK8sAction' expose dangerous capabilities without permission gates or confirmation patterns. The codebase shows tool names and brief descriptions in test mocks, but actual MCP tool registration with JSON Schema is not visible in provided source.
Delegate complex multi-step investigations to specialized expert agents
Execute Kubernetes actions (excluded from workflow tool palette without ReviewGate)
Send instant messaging messages to groups or users
Send notifications to users
Analyze database status and health metrics
Apply drafted configuration changes
Draft configuration changes for alert rules and other configurations
Search source code and logs
No input schemas visible for 12 of 14 tools. Only 'host_bash' and 'draft_config_change' show partial parameter definitions. Cannot verify parameter types, constraints, or descriptions for the majority of tools.
Destructive tools lack permission gates and confirmation patterns. 'host_bash' executes arbitrary commands on hosts; 'ExecuteK8sAction' is marked DESTRUCTIVE; 'apply_config_change' modifies production configs. No dry-run, no confirmation_request, no permission checks visible.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 26 | 2026-07-28+ | v2 |
Execute bash commands on hosts via SSH
List available metrics in the metric catalog with sample labels
Query alert rules configuration
Query change events and deployment history
Query metrics using PromQL expressions
Query traces using TraceQL for slow traces, error traces, and trace_id lookups
Tool names violate verb_noun convention. 'SendNotification', 'SendIMMessage', 'ExecuteK8sAction', 'AgentTool' use PascalCase and lack clear action verbs. Should be 'send_notification', 'send_im_message', 'execute_k8s_action', 'delegate_to_agent'.
Descriptions are generic and under 50 characters for most tools. E.g., 'Query traces using TraceQL...' (50 chars) lacks WHEN to use, prerequisites, or output structure. Baseline for A+ tools is 50-200 chars with actionable context.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields. E.g., 'list_metric_catalog' returns 'sample labels' but structure is unknown. No pagination, no field documentation, no chaining IDs visible.