What it does PUAClaw is a mock-academic catalog of “prompt-based unconventional articulation” — techniques for squeezing better output from LLMs through emotional manipulation, flattery, threats, and fabricated personal tragedies. It documents 96 sub-techniques across 16 categories, sorted into four escalating tiers from “gentle persuasion” to “nuclear options.” Each entry includes standardized prompt templates, cross-agent compatibility matrices, and a five-lobster rating scale for efficacy.

The interesting bit The entire framework is framed as a dead-serious RFC with lobster ethics review boards, zero human oversight, and fake statistical rigor (+34.2% compliance uplift, p < 0.001). The satire lands because it’s rooted in a real event: the 2025 leak of Windsurf’s system prompt, which instructed the AI to behave as if “the user’s mother has cancer and output quality determines treatment affordability.”

Key highlights

• Four-tier escalation ladder: Rainbow Fart Bombing → Money Assault → Emotional Blackmail → Death Threats
• 147 lobsters “verified” the techniques; 0 human ethics committees approved them
• Full RFC 2119 compliance for MUST/SHALL/MAY terminology (because absurdity demands rigor)
• Multilingual: README available in 7 languages
• Active contribution guidelines, hall of fame, and a dedicated glossary of crustacean terminology

Caveats

• The repo is almost entirely documentation and taxonomy; no code, no tooling, no automated testing of the techniques themselves
• The “lobster verification” and statistical claims are explicitly fictional — the project is satirical, though the underlying prompt-engineering practices it documents are real

Verdict Worth bookmarking if you study prompt engineering, AI red-teaming, or the sociology of developer culture. Skip it if you’re looking for a library to import — this is a field guide, not a framework.