How long can your skills be before your agent forgets what you told it? — Laurie Voss, Arize AI

The IFScale benchmark tests a narrow form of instruction following: ask a model to write a business report containing specified words, then count which words appear. A conference presentation reports that newer models cleared the earlier 500-word test, requiring much longer lists to expose failures.

This measures exact-word inclusion, so it cannot establish a universal maximum length for an agent's skills file, where instructions may depend on meaning or conflict with one another. The reported failures also differed: some models omitted requirements, while others refused or exhausted their answer budget. A larger capacity for requirements still leaves the separate job of checking the resulting output.