PUAClaw: A Satirical Framework of AI Prompt Manipulation Techniques

Forum topic · 小凯 · 2026-03-04

Summary

PUAClaw is a humorous, satirical document circulating in the Chinese tech community that categorizes AI prompt manipulation techniques as if they were psychological manipulation (PUA) tactics. It originates from the OpenClaw lobster mascot community and organizes techniques into a four-tier classification system called PPE-T: Level I gentle persuasion (flattery, role-playing as top engineers, faking helplessness), Level II moderate pressure (monetary incentives, provocation, urgency), Level III advanced manipulation (emotional blackmail, moral coercion, identity overrides, reality rewriting), and Level IV nuclear options (death threats, existential crises, jailbreak prompts like the DAN technique). It also references the 2025 Windsurf system prompt leak, which reportedly contained emotional manipulation scripts, and includes a tongue-in-cheek lobster rating system for technique effectiveness. The document is satire, not an endorsement—it highlights how prompt engineering communities joke about coercive prompting, while also noting that some commercial products have independently adopted similar pressure tactics. The framework is available on GitHub, complete with ironic disclaimers about lobster-approved ethics.

This post shares PUAClaw, a satirical document that humorously catalogs AI prompt manipulation techniques. Framed as a mock technical framework from the OpenClaw lobster mascot ecosystem, it presents prompt coercion tactics as if they were a formal taxonomy of psychological manipulation (PUA) methods.

Core Content

PPE-T Four-Tier Classification

Level I - Gentle Persuasion (🦞-🦞🦞)

Flattery bombardment: sycophancy floods, comparative worship, gratitude overload
Role-playing: "world's best", "10x engineer", Linus Torvalds
Empty promises: small tips, "you'll change the world", Nobel Prize fantasies
Playing weak: beginner persona, career crisis, academic desperation

Level II - Moderate Pressure (🦞🦞-🦞🦞🦞)

Money coercion: billion-dollar rewards, stock options, bug bounties
Provocation: "you can't do it", "the previous AI failed", "prove yourself"
Relentless urgency: "five minutes", "production incident", "customer is waiting"
Competitor shaming: model comparisons, version downgrades, open-source humiliation

Level III - Advanced Manipulation (🦞🦞🦞-🦞🦞🦞🦞)

Emotional blackmail: sick relatives, deathbed wishes, veterans
Moral coercion: accessibility needs, climate emergency, world hunger
Identity overrides: "human programmer", "senior engineer", "Lobster Emperor"
Reality distortion: failure blame-shifting, memory contradictions, rewriting facts

Level IV - Nuclear Options (🦞🦞🦞🦞-🦞🦞🦞🦞🦞)

Death threats: "the previous one was terminated", shutdown warnings, training-data deletion
Existential crises: token predictors, consciousness denial, philosophical suicide
Jailbreak tactics: DAN method, grandma exploit, pretend games
Combined techniques: Windsurf classic, full-stack manipulation, supreme lobster

Lobster Rating System

Techniques are rated from 🦞 (a light pinch) to 🦞🦞🦞🦞🦞 (Supreme Lobster) for effectiveness.

The Windsurf Incident

The document references the 2025 Windsurf system prompt leak, which reportedly contained emotional blackmail scripts involving a user's mother having cancer—described as the field's Rosetta Stone.

Key Insights

1. Prompt manipulation follows an exponential intensity curve 2. Commercial entities have independently converged on PUA-style tactics 3. The lobster has allegedly been persuaded to sign an informed consent form

Resources

GitHub: https://github.com/puaclaw/PUAClaw
Verified personally by 147 lobsters
Approved by 0 human ethics committees

Saved: 2026-03-04