Markdown Export
Universal system prompt
Export your blueprint as structured markdown and paste it into any AI system prompt or custom instructions field. Works with every AI platform that accepts text context — no API, no extension required.
Export format
Markdown structure
The exported markdown follows a consistent 4-layer structure that mirrors the blueprint JSON. Each layer is a level-2 heading with its modules as key-value pairs. The format is designed to be both human-readable and maximally parseable by LLMs.
# AI Identity Blueprint — Alex Chen
> Generated by Artificial ID · v3 · 2026-03-10
---
## Layer 1: Core Identity
**Name:** Alex Chen
**Role:** Senior Full-Stack Engineer
**Communication style:** Direct, precise, context-aware
**Work preferences:** Async-first, documentation-driven, deep focus blocks
---
## Layer 2: Voice & Tone
**Formality:** Semi-formal
**Verbosity:** Concise — prefer bullet points and headers over prose walls
**Humor:** Dry, occasional; never forced
**Preferred format:** Structured with headers; code blocks for all code
---
## Layer 3: Code DNA
**Primary languages:** TypeScript, Python, Rust
**Frameworks:** Next.js (frontend), FastAPI (API), Axum (systems)
**Code style:** Functional where possible, explicit over clever, no magic
**Tools:** Neovim, tmux, Cursor, Warp
**Philosophy:** Boring stack, exotic algorithms. Ship boring, optimize relentlessly.
---
## Layer 4: Knowledge Base
**Domains:** Distributed systems, developer tooling, AI/ML infrastructure
**Reasoning style:** First-principles — strip to axioms before rebuilding
**Mental models:** Second-order effects, Chesterton's fence, inversion
**Known biases:** Bias toward over-engineering; watch for gold-plating
---
*This file was exported from Artificial ID. Re-import or update via the dashboard.*One file, 4 layers
The export is a single self-contained markdown file. Layers are separated by horizontal rules for clear visual parsing.
Selective layer export
From the dashboard Export tab, choose which layers to include. Exclude Code DNA for non-technical contexts, for example.
Versioned header
Every export includes a version number and date. When you update your blueprint, re-export and replace to keep context current.
ChatGPT
Custom instructions format
ChatGPT custom instructions split into two fields: “What would you like ChatGPT to know about you?” and “How would you like ChatGPT to respond?” Paste your Core Identity and Knowledge Base into the first field, and Voice & Tone into the second.
# About me
{{NAME}} — {{ROLE}}
## Background
{{KNOWLEDGE_BASE.DOMAINS}}
## How I work
{{CORE_IDENTITY.WORK_PREFERENCES}}Formality: {{VOICE_TONE.FORMALITY}}
Verbosity: {{VOICE_TONE.VERBOSITY}}
Format: {{VOICE_TONE.PREFERRED_FORMAT}}
Humor: {{VOICE_TONE.HUMOR}}ChatGPT custom instructions character limit
Each field in ChatGPT custom instructions has a 1,500 character limit. The full blueprint export is typically 2,000–3,000 characters. Use selective layer export (Core Identity + Voice only, or Code DNA only) to stay within limits, or use the browser extension which bypasses this restriction by injecting into the system prompt directly.
Claude Projects
Structured identity for Claude
Claude Projects support a project-level system prompt and multiple uploaded knowledge files. The XML-tagged format below works particularly well with Claude's context parsing — Claude is trained to understand XML-structured system prompt data.
<identity>
<name>{{NAME}}</name>
<role>{{ROLE}}</role>
<communication_style>{{CORE_IDENTITY.COMMUNICATION_STYLE}}</communication_style>
</identity>
<voice>
<formality>{{VOICE_TONE.FORMALITY}}</formality>
<verbosity>{{VOICE_TONE.VERBOSITY}}</verbosity>
<format>{{VOICE_TONE.PREFERRED_FORMAT}}</format>
</voice>
<technical_profile>
<languages>{{CODE_DNA.LANGUAGES}}</languages>
<frameworks>{{CODE_DNA.FRAMEWORKS}}</frameworks>
<style>{{CODE_DNA.STYLE}}</style>
<philosophy>{{CODE_DNA.PHILOSOPHY}}</philosophy>
</technical_profile>
<knowledge>
<domains>{{KNOWLEDGE_BASE.DOMAINS}}</domains>
<reasoning>{{KNOWLEDGE_BASE.REASONING_STYLE}}</reasoning>
<mental_models>{{KNOWLEDGE_BASE.MENTAL_MODELS}}</mental_models>
</knowledge>Use knowledge files for depth
Upload your full markdown export as a knowledge file in the project. The system prompt stays lightweight (XML tags above) while the knowledge file provides deep context that Claude can reference when needed.
Or use MCP for live sync
Claude Desktop with MCP enabled pulls your blueprint fresh on each conversation — no manual file uploads. Consider MCP if you update your blueprint frequently.
Template variables
Build custom prompts
The dashboard Export tab offers a template mode where you write your own system prompt structure and insert blueprint values using double-brace variables. The template is resolved server-side and returned as a ready-to-paste string.
# Who you're talking to
{{NAME}} — {{ROLE}}
## How I communicate
- Formality: {{VOICE_TONE.FORMALITY}}
- Verbosity: {{VOICE_TONE.VERBOSITY}}
- Format preference: {{VOICE_TONE.PREFERRED_FORMAT}}
## Technical context
Languages: {{CODE_DNA.LANGUAGES}}
Frameworks: {{CODE_DNA.FRAMEWORKS}}
Philosophy: {{CODE_DNA.PHILOSOPHY}}
## What I know deeply
{{KNOWLEDGE_BASE.DOMAINS}}
## How I think
{{KNOWLEDGE_BASE.REASONING_STYLE}}{{NAME}}core_identity.nameAlex Chen{{ROLE}}core_identity.roleSenior Full-Stack Engineer{{VOICE_TONE.FORMALITY}}voice_tone.formalitysemi-formal{{VOICE_TONE.VERBOSITY}}voice_tone.verbosityconcise{{VOICE_TONE.PREFERRED_FORMAT}}voice_tone.preferred_formatstructured with headers{{CODE_DNA.LANGUAGES}}code_dna.languages[]TypeScript, Python, Rust{{CODE_DNA.FRAMEWORKS}}code_dna.frameworks[]Next.js, FastAPI, Axum{{CODE_DNA.PHILOSOPHY}}code_dna.philosophyBoring stack, exotic algorithms{{KNOWLEDGE_BASE.DOMAINS}}knowledge_base.domains[]Distributed systems, dev tooling{{KNOWLEDGE_BASE.REASONING_STYLE}}knowledge_base.reasoning_styleFirst-principlesToken counts
Estimated token usage
These are approximate token counts using the GPT-4 / Claude tokenizer (cl100k_base). Actual counts vary by your content length. Use selective exports to minimize context overhead.
For most coding workflows, the Code DNA layer alone (~100–180 tokens) provides the highest signal-to-token ratio. Add Core Identity (+~80 tokens) when the AI needs to know how to address and communicate with you, not just how you code.
Integration patterns
System prompt integration patterns
Different contexts call for different integration strategies. These four patterns cover the most common use cases — from maximum simplicity to maximum control.
Full identity prefix
Paste the entire exported markdown at the top of your system prompt. Best for general-purpose assistants where you want full context.
Code DNA only
Export only the Code DNA layer and paste into coding assistant system prompts. Keeps tokens low while giving the AI your full technical profile.
Layer-by-layer injection
Some systems (like Claude Projects) support multiple knowledge documents. Split your export by layer and upload each as a separate document.
Template variable substitution
Use the template format to build a custom system prompt where only specific fields are substituted. Gives you full control over the prompt structure.
Next steps
Automate beyond copy-paste
MCP Server
Skip the copy-paste entirely. MCP injects your blueprint automatically into Claude Desktop and Cursor on every conversation.
API Reference
Build automated pipelines that fetch your blueprint and inject it into any LLM call programmatically via the REST API.
Security
Understand how your blueprint is protected — even when exported as plaintext, the source document stays encrypted.
Ready to export?
Complete your interview, then find your export in the dashboard under Distribution → Markdown.