Two tools with basic schemas but significant quality gaps. execute_command has a minimal 1-param schema with adequate description. start_tunnel has 2 optional params with defaults but lacks clarity on side effects and tunnel lifecycle. Neither tool has output schema documentation. Parameter descriptions are present but generic. No error recovery guidance. Missing idempotency hints for a destructive tool (execute_command). Tool composition is weak, no batch operations, no pagination, no chaining IDs in responses.

• Add output schema documentation: execute_command should return {type: 'object', properties: {output: {type: 'string'}, exit_code: {type: 'number'}, truncated: {type: 'boolean'}}}. start_tunnel should return {tunnel_url: string, port: number, subdomain: string, expires_at: string}.
• Implement confirmation pattern for execute_command: add a 'dry_run' boolean parameter (default false) or require explicit 'confirm: true' for shell commands. Document that this tool is irreversible.
• Enhance start_tunnel description: 'Starts a public web tunnel to the VM interface on the specified port. Closes any existing tunnel. Returns a public URL valid for ~2 hours. Subdomain must be 4-63 alphanumeric characters.'
• Replace 2-second timeout with proper async handling: return immediately with a job_id, offer a get_command_status tool to poll results, or document that only short-running commands are supported.
• Add parameter constraints: subdomain should match regex ^[a-z0-9-]{4,63}$ and be documented in the schema. Port should have min/max bounds (1024-65535).