Fleet-agent-mcp has 21 tools with consistent naming (verb_noun pattern) and descriptions present for all tools. However, critical gaps exist: parameter descriptions are often generic or missing detail; input schemas lack proper JSON Schema type definitions and constraints (no enums for constrained fields like 'group', 'status', 'priority', 'category'); output schemas are undocumented; error handling is minimal with no recovery guidance. Tool descriptions average ~100 chars but lack WHEN/WHY context. The server follows a reasonable structure but falls short of production-grade quality expected for agent composition.

• Add JSON Schema output documentation to every tool. Example: query_logs returns {logs: [{source, level, message, timestamp}], count: int, total_matching: int}. This enables LLM planning and chaining.
• Convert constrained parameters to enums in schema. Replace 'group' string with enum: ['self', 'human', 'external']. Replace 'priority' with enum: ['high', 'medium', 'low']. This prevents hallucinated values.
• Expand parameter descriptions with WHEN/WHY context. E.g., 'branch (int|null): If the current workflow node has multiple branches (e.g., approval paths), pass the 0-based branch index. Call workflow_status first to see available branches. Required if branches exist, null otherwise.'
• Add error recovery guidance. E.g., 'If workflow_start fails with "workflow not found", call workflow_list to see registered workflows. If the workflow exists but is not listed, call workflow_autodiscover to register it.'
• Implement confirmation pattern for destructive ops. Add a 'confirm' parameter (bool, default false) to pulse_delete and workflow_reset. When false, return {confirmed: false, message: 'This will permanently delete X. Call again with confirm=true to proceed.'}. This prevents accidental destruction.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry