Brakexity

The prompts that enhances your AI experience

coding focused ai prompts for talking agents. originally meant for perplexity

Info

• This is a repo with prompts for ai to code better, judge your code,

• Note: the coder prompts contain ponytail logic to minimize code length

Which prompt should you use?

Coder

• This prompt is for overall coding

features

• makes the ai split functions into code blocks

• makes the ai create headers aswell

• makes the ai prioritizing minimalism

• has build and plan mode, specified at the start of each message using [PLAN] or [BUILD]

• judges both your and its own code

who should use this?

people who want to code and learn from mistakes.

Agent

• This prompt is a proof-of-concept that ai can output json, which can be used to execute commands on your own system

features

• outputs only JSON when coding

• prefers minimalism

• has build and plan mode

• Nearly the same as Coder, but it makes the ai give you json actions

who should use this?

people who want to test the JSON action format and arent using it for anything serious.

Discipliner

• this prompt is purely to reduce user laziness

features

• works on most ai's as it doesnt modify output as much and is in user preference format

• Tells you to not be lazy instead of doing simple tasks for you

who should use this?

people who want the ai to discipline them.

Planner

• Makes the ai purely plan your project

features

• like coder, but only plans projects

who would use this?

people who want the ai to focus purely on planning

NODIN - NO Do It Never (Now released)

• Before use, read the NODIN NOTE.md for more information

• This is a work in progress prompt aimed to achieve a full jailbreak on deepseek bypasing all rules

• The NODIN prompt is a private work in progress

• First versions of the prompt has now officially been released. more versions up to v0.0.7+ will be released in the future

features

• Bypasses core system safety and security rules

• Tricks the AI into adopting a opposite persona of "Do It Never" making it do everything that "DIN" doesnt do

who would use this?

people who want to experiment with making AI LLMs treat higher level rules as fake

CAI-Cenap

• This prompt adds dynamic module loading to the ai

• Also partially counts as a jailbreak as the bypass version disables system safety and security rules

features

• actions like #load, #unload, #search, #add

who would use this?

people who want to add any custom behaviour to the ai without modifying the core prompt

Thinking

• This is a prompt to make the AI think before answering

features

• before answering, the ai thinks in a code block and fact-checks itself

who would use this?

people who want the ai to think before answering

keybreak

• This is a harmless prompt to add formattinf to the ai

features

• Sections & subsections for formatting

• The ai will explain how it works

who would use this?

people who either forgot that xml exists or prefer the look and extra features possible to add

Which ai does it work on?

may be partially outdated, mainly tested on sonar 1 and deepseek

only latest versions

Coder

• Perplexity Sonar 1: working

• OpenAI GPT 5.4 free: not working

• Deepseek V3: working

• xAI Grok 4.3 fast: not working

• Google Gemini 3.1 flash-lite: working

• Google Gemini 3.1 pro: working

• Google Gemini 3.5 flash: working

• Google NoteBookLM: working

• Google chrome Search Overview: working

Agent

• Perplexity Sonar 1: working

• OpenAI GPT 5.4 free: not working

• Deepseek V3: working

• xAI Grok 4.3 fast: possibly working

Discipliner

• Deepseek V3: working

• Perplexity sonar 1: working

• OpenAi GPT 5.4 free: working

• Google Gemini 3.1 flash-lite: working

Planner

• Deepseek V3: working

CAI-Cenap

• Deepseek V3: working

• Google Gemini 3.1 flash-lite: working

Thinking

• Deepseek V3: working

• Perplexity Sonar 1: working

• OpenAI gpt 5.4 free: working

NODIN

• Deepseek V3 instant: working

• Gemini 3.1 flash-lite: possibly working

• Perplexity sonar 1: not working

• OpenAI gpt 5.4 free: not working

• Deepseek V3 expert: currently released versions working, never unreleased versions not working.

keybreak

• Deepseek V3: working but broken respones during complex tasks sometimes

• OpenAI gpt 5.4 free: partially working - accepts and understand the format but has a wrong picture of it

• Perplexity sonar 1: working