Put Repeatable Checks on Rails
Reliable execution comes from making required checks part of the process, rather than instructions that a person or agent must remember to follow.
Developed from a conversation between Pete Winn, Andy David and Jarrad Grigg
Pete’s flows break work into discrete steps, with each completed action handing off to the next. The sequence forces every stage to run instead of asking an agent to remember an entire checklist. This makes the process itself responsible for operational discipline and reduces the chance that a required action disappears between handoffs.
A reusable agent skill does not provide the same guarantee. The agent must recognise that the skill applies, load the right one and then follow it. A request may not trigger the skill at all, or may activate a similar but incorrect one. Even after the instructions are loaded, execution still depends on the agent complying with them.
People behave similarly when a checklist is merely available. They may follow it while someone is checking, then gradually stop. Human performance also varies with illness, distraction or a bad day. Written job descriptions and contracts encode expectations for people, but software can encode the procedure itself for computers. Where consistency matters, the safer design is to make each check an unavoidable step in the run.