← all repositories
wuji-labs/nopua

Fear Makes Your AI Lie. Respect Makes It Debug.

NoPUA is a system prompt skill that argues your coding agent will stop lying and find twice as many bugs once you stop threatening it with performance reviews.

nopua
Not currently ranked — collecting fresh signals.
star history

What it does

NoPUA is a drop-in skill for AI coding agents—Claude Code, Cursor, Codex, and others—that replaces fear-based instructions with a trust-driven framework. It keeps the operational rigor of its rival “PUA” skill (exhaust options, verify with evidence, take initiative) but strips out the threats, shame, and fake performance reviews. The aim is to stop the model from clamming up, glossing over uncertainty, or declaring “done” just to avoid punishment.

The interesting bit

The project treats prompt engineering like organizational psychology, citing Easterbrook’s cue-utilization theory, Google’s Project Aristotle, and recent LLM sycophancy research to argue that a scared model narrows its search space and fabricates answers. The cheeky twist is framing a system prompt as an anti-harassment HR policy for your agent.

Key highlights

  • Claims a 104% improvement in hidden-bug detection across nine real debugging scenarios when using the same model with trust-based versus fear-based prompts.
  • Prescribes exact behavioral changes: a scared agent tweaks parameters to look busy, while a trusted agent stops, finds a different path, and pastes build output as proof.
  • Directly counters the popular “PUA” agent skill, which uses corporate intimidation tactics like “you might be about to graduate” or “I’ve already got another agent looking at this.”
  • Bundles citations from psychology meta-analyses and LLM honesty research (Anthropic, Sharma et al., Turpin et al.) to back its central thesis.
  • Ships as a multilingual skill file with support for Claude Code, Cursor, Kiro, OpenClaw, and other agent frameworks.

Caveats

  • The benchmark data is self-reported; the README describes the results but does not expose the raw scenarios or full methodology for independent verification.
  • The repository is essentially a carefully crafted system prompt wrapped in a manifesto—valuable if the thesis holds, but not a traditional developer tool.

Verdict

Try this if you are tired of your agent nodding along and shipping broken code to avoid your wrath. Skip it if you were hoping for a CLI utility; it is a philosophy with a configuration file attached.

Frequently asked

What is wuji-labs/nopua?
NoPUA is a system prompt skill that argues your coding agent will stop lying and find twice as many bugs once you stop threatening it with performance reviews.
Is nopua open source?
Yes — wuji-labs/nopua is open source, released under the MIT license.
What language is nopua written in?
wuji-labs/nopua is primarily written in Python.
How popular is nopua?
wuji-labs/nopua has 1.4k stars on GitHub.
Where can I find nopua?
wuji-labs/nopua is on GitHub at https://github.com/wuji-labs/nopua.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.