11 tools with mixed quality. All have descriptions (good), but most lack depth and LLM-optimization guidance. Schemas present but minimal, parameters have types and descriptions, but lack constraints (enums, ranges, patterns). No output schemas documented. Error handling exists but is generic. Tool names follow verb_noun convention (system_info, disk_usage, service_status) but lack clarity on when to use each vs. similar tools. Approval workflow tools (approval_list, approval_get, approval_approve, approval_reject) are well-named but descriptions don't explain the approval lifecycle or prerequisites. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels (READ_ONLY vs. WRITE). write_config_file lacks confirmation/dry-run pattern for destructive operations.

• Document output schemas for all 11 tools. Include field names, types, and descriptions. Example: system_info returns {hostname: string, kernel: string, platform: string, user: string}.
• Add tool annotations: mark all READ_ONLY tools with readOnlyHint=true; mark write_config_file with destructiveHint=true; mark approval_approve and approval_reject with idempotentHint=true.
• Add constraints to parameters: top_processes.limit should be 1 - 100 (default 5); approval_list.status should be an enum [pending, executing, completed, failed, rejected].
• Implement dry-run or confirmation for write_config_file. Return a preview of changes and require explicit approval before writing, or add a 'dry_run' boolean parameter.
• Expand approval workflow descriptions to explain the lifecycle: 'Approval workflow: pending → (approve/reject) → executing → (completed/failed). Call approval_list to discover pending approvals, approval_get for details, approval_approve to execute.'