Agent RL: train a policy inside a harness

Assumptions, implementation & acceptance

Tool budgets are nonnegative integers. Allowlist checks precede execution; exhausted budgets stop the loop. This deterministic mock tests orchestration, not an LLM’s planning competence.

Formula conditions, units and worked arithmetic reviewed. The implementation isolates one numerical mechanism; it does not train a policy or control real hardware.

Before running: predict a boundary case

The third action is never executed. This tests a budget gate, not a learned agent.

Minimal executable example

Save the example and linked modules in one folder. Use Node.js 22 or newer, then run node agent-rl.mjs. No third-party runtime dependencies.

const solve=()=>{
const costs=[1,2,4], budget=3;let spent=0,steps=0;
for(const cost of costs){if(spent+cost>budget)break;spent+=cost;steps++}return [spent,steps];
};
const actual=solve();
const expected=[3,2];
if(JSON.stringify(actual)===JSON.stringify(expected)){console.log('PASS',actual)}else{
const a=[actual].flat(), b=[expected].flat();
if(a.length!==b.length||a.some((v,i)=>!Number.isFinite(v)||Math.abs(v-b[i])>1e-10))throw Error('Numerical mismatch');
console.log('PASS within 1e-10',actual);
}

[
  3,
  2
]

Test and transfer

The third action is never executed. This tests a budget gate, not a learned agent.

Record input, code version, expected result and observed result in the notebook below. Then construct a counterexample that violates the assumption above and explain why the original conclusion no longer follows.