Reverse-engineer how you think, talk, and make decisions — then turn it into a reusable AI System Prompt.
Built by WhyStrohm — embedded creative engine for founder-led brands.
| Scenario | Generic AI | With a Twin |
|---|---|---|
| Client ghosted your proposal | "I hope you're doing well! I wanted to follow up on the proposal I sent over..." | "12 days of silence after a $12K proposal means one of two things: scope mismatch or timing mismatch..." |
| Teammate suggests a worse approach | "That's an interesting idea! I can definitely see the appeal of something more straightforward..." | "That's not simpler — that's labor disguised as simplicity..." |
| Pitching a new client | "Great to connect! [Referrer] spoke really highly of you and your brand..." | "[Referrer] made the intro — so I'll skip the preamble and go straight to the structural problem..." |
→ See full before/after comparisons
This is a persona extraction workflow that analyzes your writing to build a "Digital Twin" — a System Prompt that makes any AI replicate your voice, judgment, and decision-making logic.
It's not a personality quiz. It reads how you actually communicate and extracts:
- Linguistic Fingerprint — vocabulary, metaphor style, crutch words, formatting habits
- Cognitive Pattern — how you organize information and solve problems
- Decision Logic — the rules you use (consciously or not) to say yes or no
- Knowledge Map — where you go deep vs. where you skim
Then it stress-tests the result to make sure it holds under pressure.
| Level | Tool | Data Source | Depth |
|---|---|---|---|
| Layer 1 | Any LLM (ChatGPT, Gemini, Claude, etc.) | Paste your writing samples | ~70% |
| Layer 2 | Claude with memory enabled | Pulls from conversation history | ~85% |
| Layer 3 | Claude Code or Cowork | Scans your actual local files | ~100% |
- Copy the contents of
EXTRACTION_PROMPT.md - Paste into a new chat in any LLM
- Add your writing samples at the bottom (5+ pages recommended)
- Let it run all phases including the stress test
- Open a new chat in Claude (memory must be enabled)
- Paste the contents of
EXTRACTION_PROMPT.md - No writing samples needed — Claude pulls from your conversation history
- Let it run all phases
- Create a folder with your writing:
mkdir ~/digital-twin-scan - Copy your files in (scrub sensitive details first)
- Paste the contents of
CLAUDE_CODE_PROMPT.mdinto Claude Code - Follow the 7-step pipeline — includes quantitative analysis and visual dashboard
Upload digital-twin-skill.zip to Claude.ai (Settings > Skills) or drop SKILL.md into ~/.claude/skills/digital-twin/
├── README.md ................. You're here
├── CHANGELOG.md .............. Version history
├── EXTRACTION_PROMPT.md ...... Universal prompt (any LLM)
├── CLAUDE_CODE_PROMPT.md ..... Full 7-step pipeline with file scanning
├── SKILL.md .................. Claude Code / Claude.ai skill
├── digital-twin-skill.zip .... Upload-ready for Claude.ai
├── SAFETY_CHECKLIST.md ....... Pre-paste, pre-scan, pre-share rules
├── LICENSE ................... MIT
├── assets/ ................... Demo GIF and social preview
├── examples/
│ ├── before-after/ ......... 3 side-by-side comparisons
│ └── sample-profiles/ ...... 3 complete Twin profiles
└── validation/
├── RUBRIC.md ............. 10-dimension weighted scoring
├── STRESS_TESTS.md ....... 15 adversarial prompts
└── SCORED_EXAMPLE.md ..... Worked scoring example (7.75/10)
Don't guess whether your Twin is good. Measure it.
- Scoring Rubric — 10 weighted dimensions. Voice Accuracy and Decision Consistency carry 30% of the score. Below 7.0 needs rework.
- 15 Stress Tests — adversarial prompts across 5 categories: high stakes, conflict, ambiguity, context shift, edge cases.
- Scored Example — a Founder/CEO Twin scored at 7.75/10. See exactly where it passed and where the AI tells leaked through.
A strong Twin passes 12+ of 15 stress tests with a weighted score above 7.0.
Running the full extraction on 60 files (27,342 words) produced:
Quantitative findings:
- Zero hedging instances ("I think," "maybe," "perhaps") across the entire corpus
- Dominant metaphor family: Architectural/Structural (309 instances)
- Top crutch phrase: "content infrastructure" (26x)
- Sentence structure: 46% short (punchy), 35% medium, 19% long
- 97% declarative sentences, 0% questions
The stress test:
"$50K for manual labor — prestigious but breaks every rule in your Decision Logic."
The Twin declined, explained why using the extracted logic framework, and counter-pitched a systems version. Which is exactly what the subject would have done.
The surprising pattern:
"You build measurement tools you don't use. The Content Spiral names this exact pattern in your clients. It may also be yours."
Layer 3 also generates a visual dashboard with word frequency clouds, topic cluster maps, metaphor breakdowns, and tone spectrum charts — saved as an HTML file you open in your browser.
The prompt itself doesn't collect data, call APIs, or execute code. The risk is in what you paste into it.
Rules:
- Only paste YOUR writing. Never client data or coworker messages.
- Scrub names, dollar amounts, and identifying details before pasting.
- The prompt extracts principles, not data — no names in the output.
- Share your patterns freely. Keep the System Prompt private.
- Layer 3: create a dedicated folder. Never scan your entire home directory.
- Once AI reads a file, it's in the session. Scrub BEFORE you scan.
Full checklist in SAFETY_CHECKLIST.md.
Once you have your Twin, score your published content against it:
- Content Audit — 5-layer diagnostic that scores your content and rewrites one piece live
- Voice Scorer — measures drift between your website voice and social content
- Voice Extract — extract a structured voice profile from any URL
Then run all of them on a schedule:
- Ritual — Claude Code skill + bootstrap scan that reads your shell history, git repos, and Claude Code memory, then drafts a scheduled trigger that runs Digital Twin (and the other skills) across every brand you ship. Paste-ready for Claude Code's scheduled triggers.
See CHANGELOG.md for version history.
This is one component of the full brand infrastructure I build for founder-led brands. The Digital Twin extracts the voice. Then I build the rest.
Voice guardrails encoded into every content pipeline. Programmatic video production. Automated publishing across all channels. One operator, full stack, 30 minutes of your time per week.
11 brands. 800+ videos. You own everything I build.
See client proof → whystrohm.com/results
MIT — use it, fork it, improve it.
Built by WhyStrohm — I become your brand's creative engine.