Free tool
Loop linter
Paste a loop design — yours, your agent's, or one you found on X — and get a 0–10 score against the anti-pattern rubric: verifier present and objective, explicit stop condition, budget, one change per round, durable state, the same check every round, a micro-metric objective, human gates on risky surfaces, and a defined trigger.
Everything runs in your browser. Nothing is uploaded, logged, or stored — it is the same deterministic text lint the MCP server exposes as <code>critique_loop</code>. “Missing” means “not stated”: state it and lint again.
What a solid loop states
A loop that scores 8+ answers four questions: what is it trying to accomplish, how will it know the latest attempt worked, what does it do with what it learned, and when does it finish or ask for help. Stop conditions are enumerated, not implied: success with evidence, no measurable progress, budget exhausted, blocker, or an approval boundary.
<a class="button primary" href="/library/">Steal a loop that already passes</a> <a class="button secondary" href="/for-agents/">Run this rubric from your agent (MCP)</a> <a class="button secondary" href="/ai-loop-template/">Loop template</a>