[FEATURE]: Add isReadOnly and isDestructive flags to tool interface #20296

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Add isReadOnly and isDestructive optional flags to the Tool.Def interface. These flags enable the permission system, auto-approval classifiers, and UI to make smarter decisions about tool behavior without analyzing tool internals at runtime.

Problem: Currently, the permission system and UI have no structured way to know whether a tool only reads data or can destructively modify state. This forces:

• Permission prompts for every tool call, even safe read-only ones
• No differentiation between read (safe) and write (destructive) at the system level
• Auto-approval classifiers must analyze tool behavior heuristically instead of using explicit metadata

Proposed Solution: Add two optional boolean flags to Tool.Def:

export interface Def<Parameters extends z.ZodType, M extends Metadata> {
  description: string
  parameters: Parameters
  /** Whether this tool only reads data (never modifies state). Used for permission auto-approval. */
  isReadOnly?: boolean
  /** Whether this tool can destruct or irreversibly modify state. Used for elevated permission prompts. */
  isDestructive?: boolean
  execute(args: z.infer<Parameters>, ctx: Context): Promise<{ ... }>
}

Tools to annotate:

Impact: Enables future auto-approval classifiers, smarter permission prompts, and UI differentiation between safe and destructive operations.

Risk: None — optional fields with no behavioral change. Purely additive metadata.