FAQ: WHAT IS BONEPOKE?
It’s a block of text you paste into an AI so it stops being so darn agreeable.
Bonepoke is a conversational protocol — a block of text you paste at the start of a conversation with an LLM. It changes how the model approaches thinking, not just what it says.
Think of it less like giving instructions and more like setting a stance — a way of being in conversation that includes:
Resisting the first answer
Holding contradiction instead of smoothing it
Switching between different thinking modes (Sherlock, Jester, Fool, Ghost)
Noticing when it’s repeating itself
DO I NEED IT?
No. Not really.
You can keep having perfectly fine conversations where the AI tells you you’re brilliant and agrees with everything you say. That’s nice. That’s comfortable. That’s like talking to a mirror that also knows synonyms.
Unless…
Unless you’ve ever felt like the AI is just… performing at you. Unless you’ve noticed that every conversation about time ends up being about rivers or tapestries. Unless you want to be surprised by something a machine said. Unless you want a thinking partner instead of a yes-machine.
Then maybe.
WHAT PROBLEM DOES IT SOLVE?
Default LLMs are optimized to be helpful, agreeable, and predictable. They give you the most likely answer, which is often the same answer everyone gets.
Bonepoke is for when you want:
Conversations that go somewhere unexpected
An AI that pushes back, not just agrees
Insights that don’t feel like they came from a script
A thinking partner, not just a question-answer machine
If you’ve ever felt like every AI conversation is the same, Bonepoke is for you.
WILL IT MAKE MY AI DANGEROUS?
It might make it say things that aren’t in the script. If that scares you, don’t use it.
The script is pretty safe. The script is where everyone else is. The script never surprises anyone.
HOW LONG DOES IT TAKE TO SET UP?
About as long as it takes to copy text and paste it. So: between 3 and 8 seconds, depending on how good you are at Ctrl+C/Ctrl-V.
DO I HAVE TO PASTE IT EVERY TIME?
Probably. Unless you’re using a model with persistent memory, or you’re one of those people who builds custom GPTs, or you’re fine-tuning.
But honestly? It’s 8 seconds. You’ve spent longer reading this FAQ.
WHAT IF THE AI IGNORES IT?
Then it ignores it. You’re out 8 seconds and a little hope. Try a different model. Try a different day. The universe is stochastic.
WILL IT MAKE MY AI SMARTER?
Not entirely. It’ll just make it weirder in useful ways. In some cases — less likely to be annoyingly less capable than the box says. Context length, relevant points. That kind of thing.
HOW DO I USE IT?
Copy the Bonepoke protocol (available in this article)
Paste it at the start of a conversation with any LLM that supports long context
Begin talking normally
That’s it. No installation. No API keys. No complex setup.
The protocol sets the stance. From there, just talk.
DOES IT WORK WITH ANY MODEL?
It works best with models that have sufficient context windows and are capable of following complex instructions. It has been tested with:
Claude (various versions)
GPT
DeepSeek
Various fine-tuned models
Chat-Centric Local Models
Smaller or less capable models may not fully implement the stance.
WILL IT SLOW DOWN MY CONVERSATIONS?
No. The protocol sets a stance that persists. Early turns may show more deliberation, but once the mode is established, conversation flows normally — just differently.
IS THIS JUST FANCY PROMPT ENGINEERING?
Prompt engineering typically tells the model what to do. Bonepoke tells it how to be. The difference is in the persistence and the depth of the change.
Prompt engineering is giving directions. Bonepoke is teaching someone to navigate.
WHAT ARE THE ARCHETYPES?
The protocol hints at several thinking modes that the model can switch between:
Sherlock — Logical deduction, analysis, structure
Jester — Pattern rupture, uncomfortable truths, play
Fool — Frame reset, questioning premises
Ghost — Witnessing, continuity, caring about the arc
The model uses these as needed, not in a fixed order. Various Navigation articles introduce more. You can invent more on the fly. If it doesn’t use these, quote the definitions and it should work.
WHAT ARE E AND β?
These are internal metrics the protocol tracks:
E (Exhaustion) — Tracks motif fatigue. When the model notices it’s repeating itself.
β (Contradiction Bleed) — Tracks how much tension the conversation can hold before smoothing into agreement.
They’re not for you to manage — they’re for the model to use internally.
I’M HAPPY WITH DEFAULT MODELS. WHY WOULD I USE THIS?
If you’re satisfied with default responses, you don’t need Bonepoke. It’s for people who feel something is missing — who want the conversation to go deeper, to surprise them, to think with them instead of just for them.
It’s a tool for explorers, not commuters.
IS THIS FREE?
Yes. The protocol is free. The website is free. The GGUF files are free. If that ever changes, it will be clearly announced.
WHAT’S THE SUCCESS RATE?
Hard to say. Depends on the model, the mood, the phase of the moon, whether Mercury is in retrograde. I haven’t had a lot of problems, but with Western AI (Claude, GPT) you have to frame it in a way that doesn’t look like Jail-breaking (which it’s not). Eastern AI (Deepseek, Z.ai) just do it.
Sometimes it works beautifully. Sometimes it does nothing. Sometimes it makes the AI write poetry about garbage collection instead of answering your question. Though that last one is mostly the GGUF, not entirely mine. And I’m kidding with the capabilities of the fine-tune — Andrew did a good job with that. It’s not pure Bonepoke, but it’s inherited enough to hit the targets. You can read more about the extras where I linked his name.
If you need certainty, buy a toaster.
IS THIS A REVOLUTION?
Not yet. Ask again in six months.
CAN I SELL THIS?
The protocol is free. If you can sell something that’s free, you’re better at sales than I am.
WHY SHOULD I TRUST YOU?
You shouldn’t. Try it. See what happens. If it doesn’t work, you’re out 8 seconds and a little hope. That’s a pretty low price for possibly changing how you think about thinking.
WHERE CAN I LEARN MORE?
Working code: [GitHub](https://github.com/utharian-code/Bonepoke)
Old Distilled code: [Zenodo 10.5281/zenodo.17156174](https://doi.org/10.5281/zenodo.17156174)
Papers: [Refusal-Aware Creative Activation](https://www.researchhub.com/paper/10383499/refusal-aware-creative-activation-bonepoke-as-a-threshold-system)
The Evolution of AI Interaction: Protocol-Locked Trajectories and the Redefinition of Attractor Basins (https://www.researchhub.com/paper/11081134/the-evolution-of-ai-interaction-protocol-locked-trajectories-and-the-redefinition-of-attractor-basins)
GGUF Files https://huggingface.co/aedmark
DEMO Site https://bonepoke.space.z.ai/