nopua
An agent skill that replaces fear-based prompting, such as threatening a poor performance review, with a respectful tone, reporting that a trust-based version of the same debugging methodology found twice as many hidden bugs in testing.
nopua is a prompt-engineering skill for AI coding agents that responds to a specific prior skill which used corporate fear tactics, such as threatening a low performance review, to push an agent to verify its work. nopua keeps the same underlying methodology, such as exhausting options, verifying work by running the build, and searching before asking, but reframes it around respect and trust instead of fear.
The authors report a benchmark comparing the same model on nine real debugging scenarios under both prompting styles, in which the trust-driven version found 51 more production-critical hidden bugs than the fear-driven one, a 104% increase, and cite psychology research on how fear and threat narrow attentional focus. It is compatible with Claude Code, Codex, Cursor, Kiro and OpenClaw, and is written in Python under the MIT license.