What is a Cooldown Period?
A mandatory waiting period imposed after a policy violation or rate limit hit before the agent can retry the tool call. Cooldowns prevent rapid retry loops that waste resources and amplify the impact of policy violations.
WHY IT MATTERS
When an AI agent's tool call is denied by a policy, most agent frameworks immediately retry. Without a cooldown, the agent enters a tight loop — denied, retry, denied, retry — consuming proxy resources, flooding audit logs, and potentially triggering alerts. The cooldown period breaks this cycle by enforcing a minimum wait before the next attempt.
Cooldowns are especially important after security-related denials. If an agent attempts a forbidden operation, immediate retries suggest the agent may be in a failure mode — hallucinating actions, misinterpreting instructions, or under adversarial influence. A cooldown gives the agent time to receive the denial, process it, and ideally choose a different approach.
The length of the cooldown should match the severity of the violation. A rate limit hit might warrant a 5-second cooldown. A forbidden operation might warrant 60 seconds or more. Escalating cooldowns — doubling the wait on each consecutive violation — are particularly effective at containing persistent misbehaviour without permanently blocking legitimate agents.
HOW POLICYLAYER USES THIS
Intercept supports cooldown periods in YAML policies, configurable per tool or per violation type. When a policy violation triggers a cooldown, Intercept tracks the timestamp and rejects subsequent calls to the same tool until the cooldown expires. The denial response includes the remaining cooldown duration, allowing well-behaved agents to wait appropriately.