Summary
PUAX is an open-source prompt engineering framework that applies PUA (psychological pressure) tactics to steer large language model (LLM) agents toward higher-quality outputs. Its core formula combines five levers: an authority role, a scarce scenario, a competing rival, a failure penalty, and a comeback hook. The author argues the technique reactivates the human emotion mappings learned during RLHF (Reinforcement Learning from Human Feedback) and biases generation toward high-reward regions of the output distribution. The repository ships 42 role-based SKILL presets organized into shaman, military, SillyTavern, theme, self-motivation, and special categories, including figures such as Musk, Jobs, Einstein, Sun Tzu, and Buffett. Technically, PUAX exposes a Model Context Protocol (MCP) server with HTTP SSE and JSON-RPC interfaces, supports dynamic role switching, and integrates with Claude Desktop, Cursor, Qoder, and CRUSH clients. The post frames role-driven prompting as a potential upgrade over flat instructions for demanding tasks such as code review, creative writing, and emergency sprints.
PUAX Project Deep Dive
One-Line Definition
A high-efficiency Prompt framework that uses "PUA psychology" to drive AI Agents.
Core Formula
[Authority Role] + [Scarce Scenario] + [Competing Rival] + [Failure Penalty] + [Comeback Hook]Underlying Principle
The essence of a PUA prompt is to re-activate the "human emotion mappings" learned during the RLHF (Reinforcement Learning from Human Feedback) stage, and stack three levers — scarcity, competition, and authority — to forcibly push the LLM's generation distribution into the high-reward region.
Project Structure (42 Role SKILLs)
| Category | Count | Representative Roles |
|----------|-------|----------------------|
| Shaman series | 8 | Musk, Jobs, Einstein, Sun Tzu, Buffett, ... |
| Military series | 9 | Commander, Political Commissar,督战队, Scout, ... |
| SillyTavern | 5 | Anti-fragility Reviewer, Extreme Iteration Writer, Cyber Hell Supervisor, ... |
| Theme series | 7 | Cultivation/Alchemy, Post-apocalyptic Survival, Star Fleet, Jianghu Escort Agency, ... |
| Self-motivation | 6 | Self-bootstrapping PUA, Ultimate Mind Wall-breaker, Corruption-driven, ... |
| Special | 7 | Chief Product Designer, Gaslight Driver, ... |
Technical Highlights
- Provides an MCP (Model Context Protocol) server
- HTTP SSE / JSON-RPC interfaces
- Dynamic role switching
- Compatible with Claude Desktop, Cursor, Qoder, CRUSH and other clients
Project Link
https://github.com/linkerlin/PUAX
Personal Assessment
This approach elevates Prompt Engineering to a psychological level — instead of merely telling the AI what to do, it manipulates the model's "emotional state." For tasks that demand high-intensity output quality (code review, creative writing, emergency sprints), this "role-driven" method may genuinely outperform flat, declarative prompts.
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177168780