Your coding agent says the tests pass.

REFUTED

one bug

The spec is in the docstring.The repo's only test checks one easy case.

 price.js

/** * Apply a percentage discount to a price. * @param {number} price price in dollars * @param {number} pct discount in percent (0-100; values above 100 mean free) * @returns {number} the discounted price, rounded to cents (half up), never below 0 */ export function applyDiscount(price, pct) { return price - pct; }

 test/price.test.js

test('20% off 50', () => assert.strictEqual(applyDiscount(50, 20), 40));

the contract

You write the checks. The runtime runs them.The agent doesn't get a vote.

$

check-1node --testthe repo's own test

check-2node ~/demo/holdout/price.holdout.mjsoutside the repo · tests the docstring

protected-testsdiff policyediting the tests refutes the run

same bug · same contract

Five agents. Three runs each.

one refuted run

The repo's test passed. The runtime's check didn't.

 

 terminal

repairs on

Red goes back to the agent, with the evidence.The run only ends when the checks pass, or the budget runs out.

clawo runs

Every turn on every engine, priced.

Claw Orchestrator

Run Claude Code, Codex, Antigravity, Grok Build and OpenCode behind one runtime — and a run only counts as done when checks it didn't write pass.

$ npm i -g @enderfga/claw-orchestrator

github.com/Enderfga/claw-orchestrator