What's this about?
A command added to the same skill that finds prompting anti-patterns which actively hobble frontier models. Three named: verification rituals — instructions like "double-check your work" or "verify twice before responding", which frontier models take literally and which waste tokens; thoroughness and emphasis boosters — "be maximally thorough", "CRITICAL: YOU MUST ALWAYS…" — which produce verbosity and extra tool calls; and mandatory procedures and scratchpad scaffolds, fixed step-by-step structures the model no longer needs.
Why did these stop working?
Because they were compensating for weaknesses that have largely been trained out. Older models genuinely did skip verification and needed to be told to think step by step, so the community built up a vocabulary of nudges. Newer reasoning models already do all of it internally — so the instruction is no longer a nudge, it is an extra literal requirement layered on top of behaviour that was already happening. You pay tokens for the duplication and you constrain a process that was working.
Why does this matter to you?
- Most people's prompt libraries were written against models two generations old and have never been audited.
- "Taken literally" is the mechanism to internalise — a capable model does what you said, including the parts you meant as encouragement.
- It is a rare piece of prompting advice that is subtractive.