agentic-feedback-loop

@jayhawkrules

Use when Claude is being asked to build, fix, or polish something the user cares about getting RIGHT (UI matching a mock, fixing a flaky test, getting a Lighthouse score up, hitting visual parity with a competitor). Sets up a verifier loop so Claude self-corrects instead of one-shotting. The pattern - give Claude a tool that produces a checkable signal (Vitest watch, Puppeteer screenshot diff, Lighthouse score, type-checker output, axe-core report), have Claude iterate until the signal goes green, instead of "I think this looks right." Boris Cherny - "give it a way to see its result, and it'll iterate and get better." Includes specific recipes per verifier type, iteration budgets, and "when to stop iterating" rules so it doesn't spin forever. Keywords - feedback loop, iterate, self-correct, verifier, screenshot diff, Puppeteer iteration, Playwright iteration, Lighthouse loop, Vitest watch, axe-core, visual regression, agentic, polish, match the mock, get the score up.