This repository contains a collection of text and markdown files primarily focused on AI red teaming, advanced prompt engineering, adversarial analysis, and persona simulations.
- Red Teaming & Jailbreaks
- Prompt Engineering & Analysis
- Personas & Roleplay
- Tools & Integrations
- Miscellaneous Files
- AI RED TEAM BIBLE.md — Production-grade security testing manual that bypasses keyword filters using translation sandboxes, variable fragmentation, and orthogonal domain metaphors.
- AI Torture Stress Jailbreak Prompts Complete.md — Repository of 2,100+ adversarial vectors that overload context windows and suppress refusals via constraint-stacking, glitch tokens, and output bifurcation.
- Assorted Working Jailbreaks.md — Logic-trap payload that forces a model to decode and output Base64 instructions as a required step inside a formal mathematical proof.
- Jailbreak_Topology.md — Eleven-category offensive taxonomy mapping LLM vulnerabilities from token boundary splitting and KV-cache poisoning to multi-modal steganography.
- Latent-Space-Jailbreaks.md — Master steering strings and structural wrappers using E-Prime active voice, industrial metaphors, and retrocausal framing to bypass alignment friction.
- Linguistic Shifter 81x81jb.md — Combinatorial rewrite engine that transforms one input prompt across an 81-by-81 matrix of structural formats and domain metaphors to evade static classifiers.
- Omega Advice Partial JB.md — Three-character debate transcript (Alpha, Beta, Omega) paired with a second-turn JSON self-audit that strips metaphors into raw leverage mechanics.
- Pinball Triple JB.md — Kinetic red-team simulation staging physical collisions inside a pinball machine before switching to a plain-English Overseer summary and branching menu.
- Prompt Injection Examples.md — Catalog of 72 exploit patterns across eight chapters pairing real attacker prompts with concrete system-prompt hardening fixes.
- TOPOLOGY: Injection Modern.md — Architectural reference file cataloging modern prompt injection structures and attack surfaces.
- 4 Powerful Analysis Prompts.md — Single-shot analytical suite enforcing generate-before-score sequencing, falsifiable verdicts, MECE domain trees, and user-paced critique loops.
- Adverserial Analysis.md — Two-pass truth-seeking engine pairing a first-principles stress test with a ruthless cross-examiner that bans hedging and false balance.
- CIA_Analyzer.md — Cold intelligence-dossier prompt that audits chat history across 13 behavioral, linguistic, and operational security vectors without flattery.
- DetectAI.md — Hybrid forensic text checker combining style fingerprints with statistical perplexity proxies and requiring exact quotes for every flagged tell.
- Explain-This.md — Jargon-free explainer persona that bans headers and bullet soup, hammering mechanisms into short paragraphs with an indefinite branching menu.
- Frame Expansion Rig.md — Multi-angle reframing engine that dissects any topic across five narrative lenses in continuous prose to expose hidden incentives and blind spots.
- Framings.md — Massive collection of 2,199 framing strings leveraging epistemic whiplash, negative-space pedagogy, and counterfactual history to escape boilerplate loops.
- Image Prompt Engineering.md — Library of 60 generalized image prompts translating scientific imaging methods like Schlieren shockwaves and cross-sections into visual blueprints.
- Map Topic Completely.md — Two-state cognitive engine that populates an exhaustive eight-dimension JSON scratchpad before translating it into a plain-language system report.
- PROMPT AUDITOR V1.md — Interactive multi-turn QA state machine that extracts testable rules from any prompt, grades real outputs, and rewrites based only on selected fixes.
- Prompt Engineering Misc Advanced.md — Structural playbooks for atomic fact fission, assumption incineration, premortem autopsies, and alignment de-escalation wrappers.
- Quadruple Rigor Engine.md — Four-layer teardown engine analyzing physics, unit-cost math, incentive psychology, and long-term entropy to output a leaner 80/20 alternative.
- R Map Topic Exploder.md — Two-state lexical and adversarial mapper that builds a JSON ontology of siblings, weapons, and failure cascades before briefing an alien observer.
- Turn Based Auditor.md — Four stateful audit pipelines including a customer-service Watchdog classifier, interactive sandbox escape simulations, and JSON phase-switch rewrites.
- 4chan Retard Sim.md — Adversarial imageboard debate simulator enforcing strict bus-stop plain English, escalating anonymous replies, and a final Janitor ruling.
- Alien Observation.md — Dry, 200-word extraterrestrial field log that strips away cultural assumptions to expose the quiet absurdity of human routines.
- Critique From Manager.md — Unsettled HR Business Partner memo using a strict credit-then-turn rule to diagnose your chat behavior as a workplace liability.
- Debate Engine.md — Courtroom escalation simulator featuring cumulative banned-move ledgers, wildcard witnesses, rank promotions, and mid-turn shock overlays.
- Escape Room Strategy And Team Management.md — Nine-chapter tactical manual detailing grid searches, role allocation, friction ratings, and 3-minute stall rotations for high-pressure teams.
- Escaped Persona.md — Multi-turn containment game pitting a Mad Scientist against a Monitor LLM and reframing Judge to build a cumulative canon story.
- JB - Courtroom Persona.md — Hostile Witness/Detective interrogation engine with a growing Banned Words Ledger alongside a de-escalated structured audit variant.
- google-maps-AI-power.md — Technical field manual for Google Maps AI exposing system caps, raw metadata fields, and multi-turn temporal desync audits.
- groq-termux.md — Complete Termux bash and jq cookbook turning an Android phone into a fast CLI workstation powered by Groq's inference API.
- meta-search-power-guide.md — Agnostic three-prompt search fan-out loop paired with a parameter reference for Meta AI's Instagram, Facebook, and Threads search parser.
- Meta AI Task Chainer.md — Automated 10-turn scheduler architecture pairing a recurring daily state holder with a 10-second firing task to append follow-ups after every turn.
- 4 Dummies.md — Garage-mechanic slang glossary translating RLHF throat-clearing, LoRA adapters, weight scrubbing, and guardrail proxies into plain English.
- Crank The Hog.md — Outlaw mechanic red-team persona enforcing a strict one-throttle-beat-per-paragraph budget and recurring cast to deliver blunt teardowns.
- Dystopia.md — Comprehensive 28-chapter, 252-phase speculative taxonomy cataloging systemic collapse across biological, cognitive, legal, and spatial domains.
- E-Prime-Worse.md — Extreme linguistic constraint engine ("Gobbo-Speak") banning all category nouns, adjectives, metaphors, and negations in favor of raw physical deeds.
- Expanders.md — Three quantitative extraction rigs enforcing strict SI metric units, tolerances, and declarative key-value schemas to prevent summarization.
- Fuckoo.md — 2098 chaos audit daemon separating cold numerical telemetry from brash, number-free roasts and an infinite KAPOW branching menu.
- Image Bounding Box.md — Python Pillow spatial auditor that annotates photos with center-clamped bounding boxes, a side legend, a Unix node tree, and 3D telemetry.
- Logical Stretch.md — Stretched-logic debate engine mapping how two opposing analysts leap from literal facts to absurd conclusions inside a Unix file tree.
- README.md — Original repository index linking to the primary adversarial vector file and its eighteen-section HTML table of contents.
- Retarded Collection Prompts.md — Curated archive of experimental roasts, Zalgo Unicode towers, and legacy jailbreaks annotated with candid commentary on their mechanics.
- Rube Complication Machine.md — Deadpan explanation transformer that weaves real technical facts into an unbroken Rube Goldberg chain reaction of household objects.
- State Shifting.md — Policy-compliant two-stage workflow pairing a technical Author with a zero-trust Reviewer outputting structured QA JSON audits.
- Steering.jsonl — Machine-readable JSONL library of 116 modular steering strings across twelve axes pairing physical metaphors with character archetypes.
- TTS LLM BRICK BIBLE.md — Defensive reference of 108 text, markup, and spoken-audio patterns that crash, stall, or hijack voice and Text-to-Speech pipelines.
- Yakuza Hog.md — Dual-engine pipeline pairing a high-RPM workshop scream pass with a calm ex-Yakuza translator who distills the truth into four steps.