Server has 8 tools with basic descriptions and no visible input schemas in the provided code. Tool names follow verb_noun convention (probe_subsystem, get_status, list_subsystems, run_audit, smoke, results, score, request_effect), which is good. However, most descriptions are under 100 characters and lack actionable context. Only 3 tools show explicit parameter definitions (request_effect has action/target/reason with descriptions). The remaining 5 tools either have no parameters or their schemas are not visible in the provided source. Error handling is absent, no guidance on what to do if a probe fails or a subsystem is unavailable. Output schemas are not documented. This is a typical community-grade server with minimal polish.

• Add explicit input schemas to all 7 tools lacking them. For probe_subsystem, define: {"name": {"type": "string", "description": "The subsystem name (e.g., 'auth', 'database', 'cache'). Call list_subsystems() first to see available names."}}. For get_status, smoke, results, score: define empty input schema {}.
• Document output schemas in tool descriptions. E.g., 'Returns {"status": "healthy|degraded|offline", "last_check": "ISO8601", "details": "string"}'. This lets LLMs know what fields to extract and chain to downstream tools.
• Expand descriptions to 100-200 characters with actionable context. E.g., 'Probe a single subsystem by name and return its health status (healthy/degraded/offline). Use this to check one component; call get_status() to probe all at once. Returns last_check timestamp and error details if offline.'
• Add error handling guidance. E.g., 'If the subsystem is offline, the tool returns status=offline with error details. Retry after 30 seconds or escalate to ops. If the subsystem name is invalid, call list_subsystems() to see available names.'
• Add confirmation/dry-run pattern to run_audit and request_effect. E.g., add a 'dry_run' boolean parameter (default false) so agents can preview the audit or request before committing. Or require explicit user confirmation via a separate confirm_audit() tool.