Policy Learner Kit

Automatically adjust agent policies based on accumulated outcome data.

Configuration

Learning rate

Require human approval

Total

Improved

Degraded

Policy Evolution

0/4 learned

Retry Policypending

Before: max_retries=3, backoff=exponential

After: max_retries=5, backoff=linear for network errors

Effectiveness

72%→89%

Rate Limit Policypending

Before: 100 req/min per client

After: dynamic: 50-200 req/min based on client tier

Effectiveness

65%→84%

Timeout Policypending

Before: global timeout: 30s

After: per-endpoint: 5s-45s adaptive

Effectiveness

58%→81%

Fallback Policypending

Before: return cached data on error

After: cascade: cache > stale > default response

Effectiveness

70%→91%

Integration Code

import { createPolicyLearner } from 'agent-tools-kit/learning'

const learner = createPolicyLearner({ rate: 'conservative' })

const result = await learner.train(outcomeHistory)
// result.policies — array of updated policies
// result.improvements — effectiveness delta per policy