[FEATURE]: Document infinite subagent recursion risk when permission.task is set globally #17722

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

The problem: Adding "task": "allow" to the global permission block silently enables infinite subagent recursion. There's no depth limit in the codebase — subagents spawn subagents indefinitely, each with its own steps counter and context window. I hit 10 levels of nesting and 3+ hours of runaway token usage before noticing.

Why it's a trap: The default "*": "allow" wildcard does NOT trigger this — only an explicit "task" key does. So adding "task": "allow" alongside other permissions like "bash": "allow" seems harmless but completely overrides the built-in nesting guard in task.ts:

const hasTaskPermission = agent.permission.some((rule) => rule.permission === "task")

Neither steps (per-session counter, each child gets its own) nor doom_loop (single-session detection) protect against this.

The ask: Add warnings to the docs:

1. Agents docs (permission.task section): Warn that global/subagent-level task permissions enable unlimited recursive nesting. Show safe vs unsafe patterns.
2. Permissions docs (task entry): Note that global "task": "allow" propagates to all agents including subagents.
3. Config docs (permissions section): Brief caution about "task": "allow" being different from other permissions.

I'm happy to submit a docs PR for this — just want to confirm the approach is welcome first.

Related issues:

• [FEATURE]: Allow configurable subagent-to-subagent task delegation with call limits #7296 — configurable subagent-to-subagent task delegation (same underlying mechanism)
• Subagents don’t respect "*": "allow" agent permissions #12566 — subagents don't respect "*": "allow" agent permissions (related permission propagation)