Idempotency-Aware Retry Safety Checkmedium

Idempotency-Aware Retry Safety Check

medium Agents in Production 15 min

NumPy allowed · plain Python earns 10 bonus points

Implement a real retry-safety check combining a tool's declared idempotency with the actual failure mode, deciding whether an agent orchestrator may automatically retry a failed tool call.

Task

Implement is_safe_to_retry(annotations, failure_reason). Retry is safe if EITHER: (a) annotations.get('idempotentHint', False) is True (repeating it has no additional effect), OR (b) failure_reason == 'timeout' AND annotations.get('readOnlyHint', False) is True (a read that may not have even reached the server is safe to redo). Otherwise return False.

Function Signature

is_safe_to_retry(annotations: dict, failure_reason: str) -> bool

Examples

Example 1: idempotent tool always safe
Input: annotations = {'idempotentHint': True}
failure_reason = 'server_error'
Output: True

Example 2: read-only tool safe on timeout
Input: annotations = {'readOnlyHint': True}
failure_reason = 'timeout'
Output: True

Example 3: read-only tool unsafe on non-timeout failure
Input: annotations = {'readOnlyHint': True}
failure_reason = 'server_error'
Output: False

Constraints

failure_reason is a string

Python3Saved ✓

def is_safe_to_retry(annotations, failure_reason):

# Your implementation here

pass

Case 1: idempotent tool always safe

Input: annotations = {'idempotentHint': True}
failure_reason = 'server_error'

Expected: True

Case 2: read-only tool safe on timeout

Input: annotations = {'readOnlyHint': True}
failure_reason = 'timeout'

Expected: True

Case 3: read-only tool unsafe on non-timeout failure

Input: annotations = {'readOnlyHint': True}
failure_reason = 'server_error'

Expected: False