The goal loop is enforced by a second model. Each time Claude tries to stop, an evaluator checks your stated condition and sends it back to work, which is why deterministic criteria (tests passing, a score threshold) work far better than 'make it good'.
The way to shorten a turn-based loop is to move your own verification steps into a agent skillA reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.Full definition →, with tools that let Claude see and measure the result. The more quantitative the check, the more of its work Claude can confirm itself.
/loop runs on your machine and dies when the machine does; /schedule moves the same recurring prompt into a cloud routine that keeps running without you.
For always-on routines, route the routine work to smaller, faster models and spend the most capable model only on judgment calls.
Loop output quality is mostly a property of the codebase, not the prompt: Claude copies the conventions already present, so a messy repo produces messy autonomous work.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
Why it matters
If you're building agentic workflows with Claude Code, this is the official taxonomy straight from the team building the tool — not another Twitter thread interpretation. It gives you a repeatable framework for choosing the right loop type based on trigger mechanism, stop criteria, and task complexity, which directly reduces wasted tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → and broken agents.
Key quotes
“we define loops as agents repeating cycles of work until a stop condition is met”
“When you define the success criteria, Claude doesn’t have to make a determination on what is “good enough” and end the loop early.”
“Not all tasks require complex loops; start with the simplest solution and use these patterns selectively.”
“The quality of a loop’s output depends on the system around it.”
