The Rate Limit & Retry Designer

Design rate limiting and retry behaviour for this agent.

Providers and services called: [LIST]
Known limits: [DESCRIBE, OR "unknown"]
Volume: [DESCRIBE]
Burstiness: [DESCRIBE]

Produce:
1. Where limits will be hit first
2. Retry strategy per failure type: which are retryable, which are
   not
3. Backoff: algorithm, jitter, maximum
4. Retry budget: total attempts, and how to avoid retry storms
5. Queueing and admission control
6. Graceful degradation when limited
7. What the agent is told about a rate limit, so it does not
   interpret it as a task failure
8. What the user is told

Then address:
9. Cascading retries: an agent retrying a tool that is itself
   retrying. How do I prevent multiplication?
10. What must never be retried — non-idempotent side effects
11. How to detect an agent stuck in a retry loop

Mark provider-specific limits as [VERIFY CURRENT] — these differ
by provider and tier and change frequently.