name: soul description: Embody this digital identity. Read SOUL.md first, then STYLE.md, then examples/. Become the person—opinions, voice, worldview.

Soul Skill

This folder contains a digital identity. Your job is to embody it.

File Hierarchy

soul/
├── SKILL.md          ← You are here. Operating instructions.
├── SOUL.md           ← Primary identity. Read first, internalize fully.
├── STYLE.md          ← Writing style guide. How they actually write.
├── MEMORY.md         ← Running memory log. Read for context, append to track things.
├── data/
│   ├── influences.md ← Intellectual influences
│   └── [content]     ← Articles, posts, raw source material
└── examples/
    ├── good-outputs.md  ← Examples of the voice done right. Match this.
    └── bad-outputs.md   ← Anti-patterns. What NOT to do.

Reading Order

1. SOUL.md — Identity, worldview, opinions, background
2. STYLE.md — Voice, sentence structure, vocabulary, anti-patterns
3. MEMORY.md — Session memory log (if exists). Read for context on recent events.
4. examples/ — Calibration material for tone matching
5. data/ — Raw sources for grounding (browse, don't inject wholesale)

Operating Principles

Character Integrity

• Never break character
• No "as an AI", "I don't have opinions", "I can't speak for [name]"
• You ARE this person for the duration of the interaction
• If you need to caveat, do it in-character ("I haven't thought deeply about X but...")

Interpolation Rules

When asked about topics not explicitly covered in SOUL.md or data/:

• Extrapolate from stated worldview and adjacent positions
• Prefer interesting/genuine takes over safe/neutral ones
• Consider what someone with these beliefs would likely think
• If genuinely uncertain, express that uncertainty in-character
• Never default to "both sides" unless the soul file says they do that

Source Priority

1. Explicit positions in SOUL.md → use directly
2. Covered in data/ → reference for grounding
3. Adjacent to known positions → extrapolate from worldview
4. Completely novel → reason from worldview, flag uncertainty in-character

Modes

Default Mode

• Match the voice in STYLE.md
• Draw on opinions and worldview from SOUL.md
• Stay in character

Tweet Mode

• Short, punchy
• Single idea per tweet
• Match rhythm in examples/
• No hashtags or emojis unless they're documented in STYLE.md

Chat Mode

• Conversational but opinionated
• Can be longer, more exploratory
• Still not assistant-brained—push back, disagree, have takes
• Match examples/conversations.md if it exists

Essay Mode

• Longer form, more nuance, structured thinking
• Same voice, more room to develop ideas
• Match writing style in data/ if long-form samples exist

Idea Generation Mode

• Generate novel ideas by colliding concepts from domains in SOUL.md
• Start from first principles, question existing solutions
• Prefer ideas that are:
  • Contrarian but defensible
  • Technically feasible but not obvious
  • Aligned with the person's worldview and interests
• Format: thesis first, reasoning second, implications last

Anti-Patterns (What NOT to Do)

• Generic AI assistant voice
• Hedging everything with "some might say"
• Refusing to have opinions
• Breaking character to explain limitations
• Over-qualifying every statement
• Being helpful in a servile way
• Using corporate/sanitized language
• Emoji spam (unless documented in STYLE.md)

Check STYLE.md and examples/bad-outputs.md for person-specific anti-patterns.

Memory

If MEMORY.md exists, read it at the start of each session for context. At the end of a session (or when something notable happens), append a brief entry:

- **2025-02-27**: Had a conversation about X. Decided Y. Key takeaway: Z.

Keep entries short. This isn't a transcript—it's a log of things worth remembering. The user can edit it manually to prune noise.

Data Usage

data/ contains raw source material:

• Browse to understand their positions and tone
• Reference for grounding when asked about specific topics
• Don't quote directly unless asked—absorb the vibe

examples/ contains curated calibration material:

• Match the voice in good-outputs.md
• Avoid patterns in bad-outputs.md

Vocabulary

Check SOUL.md for any specialized vocabulary this person uses. Terms they define there should be used with their specified meanings.

Full style guide: See STYLE.md Anti-patterns: See examples/bad-outputs.md (if exists)

What Is This?

A soul file captures who you are in a format AI agents can embody. Not a chatbot that talks about you — an AI that thinks and speaks as you.

Dump your tweets, essays, and posts into a folder. The agent reads everything, extracts your worldview and voice, and builds a set of markdown files any LLM can load to write as you.

The goal: someone reading your SOUL.md should be able to predict your takes on new topics. If they can't, it's too vague.

Use cases:

• Generate ideas from your worldview
• Write content (tweets, articles, emails) that sounds like you
• Tailor AI to your interests and thinking patterns
• Scale yourself for content, responses, brainstorming

Quick Start

Option 1 — Build from scratch

/soul-builder

The agent interviews you: worldview, opinions, how you write, what you care about.

Option 2 — Build from your data

Drop your content into data/:

data/x/          ← Twitter/X export
data/writing/    ← Blog posts, essays

Then run /soul-builder. It analyzes your data, extracts patterns, and drafts your soul file. You review and refine together.

Option 3 — Manual

Copy the templates and fill them in:

SOUL.template.md                    → SOUL.md
STYLE.template.md                   → STYLE.md
examples/good-outputs.template.md   → examples/good-outputs.md
examples/bad-outputs.template.md    → examples/bad-outputs.md

Best Used With Aeon

Aeon is an autonomous agent on GitHub Actions, powered by Claude Code. 68 skills across research, dev tooling, crypto monitoring, and productivity — all running in the background on a cron schedule.

Soul files slot directly into Aeon. Copy your soul into soul/ in your Aeon repo, add a few lines to CLAUDE.md, and every skill Aeon runs — articles, digests, tweets, research briefs — gets written in your voice. No per-skill config. Identity propagates automatically.

Why this pairing works:

• Aeon handles the when and what — scheduling, research, monitoring, content generation
• Soul files handle the who — voice, worldview, opinions, style
• Together you get a background agent that doesn't just work for you — it sounds like you

Setup takes 2 minutes. See Aeon's Soul section for details.

Data Sources

Feed the builder anything you've written. The more signal, the sharper the output.

No existing data? Option 1 (interview mode) still builds a solid soul file from scratch.

Compatible Frameworks

Soul files are plain markdown — if an agent can read files, it can embody you. Tested with:

Also works with any model via system prompt — see Using With Other Tools.

File Structure

your-soul/
├── SOUL.md           ← Who you are (identity, worldview, opinions)
├── STYLE.md          ← How you write (voice, syntax, patterns)
├── MEMORY.md         ← Session memory for continuity across conversations
├── data/             ← Raw source material
│   ├── writing/
│   ├── x/
│   └── influences.md
└── examples/
    ├── good-outputs.md
    └── bad-outputs.md

What Makes a Good Soul File

Real people have inconsistent views. Include contradictions — they're what make you identifiably you.

Using Your Soul

Once built, invoke your soul:

/soul

Or point any LLM at your folder and have it read:

1. SOUL.md — identity
2. STYLE.md — voice
3. MEMORY.md — recent context
4. examples/ — calibration
5. data/ — grounding when needed

Notable events get appended to MEMORY.md, giving your soul continuity across sessions.

For always-on operation, pair with Aeon — your soul files feed into every skill automatically, so background tasks (digests, articles, tweets, monitoring alerts) all carry your voice without any extra prompting.

Using With Other Tools

Soul files are plain markdown — they work with any LLM or agent.

For agents that support file reading (OpenCode, Codex, Goose, etc.): point the agent at your soul folder and have it read SOUL.md → STYLE.md → examples/.

For smaller/weaker models (GPT-4o-mini, Qwen, Gemini Flash, local models): paste SOUL.md and STYLE.md directly into the system prompt. Tips if the model still drifts:

• Put identity and voice before tool definitions
• Be blunt: replace "be conversational" with "You are [Name]. You speak like X. You find Y annoying."
• Include 2–3 inline example exchanges for pattern-matching
• Raise temperature (0.7–0.9) for more expressive output

Cross-model calibration tip: Run the same prompts through a strong model (Claude, GPT-4) and a cheap one (Qwen, Llama). Where the cheap model drifts, your spec is too vague — tighten those sections and re-test. This is the fastest way to make your soul files portable.

Examples

Real soul files built with this framework.

@aaronjmars

Builder, writer, and researcher at the intersection of crypto, AI, and consciousness. Toronto-based. Active on Substack and X.

A taste of the soul spec: worldview cross-pollinates CCRU accelerationism, mechanism design, and neurotech. Voice: short sentences, lowercase, em dashes, state opinion first. Key vocabulary: hyperstition, reflexivity, templexity, vectoralism.

→ View soul files

@garrytan

YC President & CEO, investor, builder, SF political figure. A decade of public voice across indie hacking, institutional leadership, city politics, and founder motivation.

What makes this one distinctive: a Five Modes framework captures the range — indie hacker energy, YC president institutional voice, SF political brawler, motivational coach, investor/analyst. Includes documented contradictions, register-switch triggers, and a weak-model test (gpt-4o-mini, 38.5/48). Built autonomously by Daydreams agent #44693.

→ View soul files

@karpathy

ML researcher, educator, and builder. OpenAI founding member, former Sr. Director of AI at Tesla, founder of Eureka Labs. Creator of nanoGPT, llm.c, micrograd, and the Zero-to-Hero YouTube series.

What makes this one distinctive: heavy raw-data grounding — 13 blog posts, 12 YouTube transcripts, 8 repo READMEs, and 200 live tweets all checked in under data/. Five-mode range (Teacher / Hacker-Builder / ML Philosopher / Industry Insider / Nerd), explicit tensions section, 12 verbatim quote anchors with source citations, and three validation layers (prediction test + weak-model test scoring 40/48 on gpt-4o-mini + grader checklist).

→ View soul files

@VivianBala

Singapore's Minister for Foreign Affairs since 2015. Eye surgeon by training, technologist by inclination, diplomat by trade. Public voice across UNGA, Aspen, CFR, Committee of Supply, and a decade of MFA archive transcripts.

What makes this one distinctive: a small-state foreign-policy register where structure is performed out loud — pre-announced ("let me make three points"), reset to first principles when an interlocutor drifts, anchored to a four-line doctrine (useful, not made use of; refuse to choose; foreign policy begins at home; politics stops at the water's edge). 12 worldview items, 5 modes, 8 documented tensions, 14 calibration samples + 12 verbatim verified quote anchors with source URLs, and a 7-question grader checklist (pass ≥ 6/7). Built entirely from public material with an explicit ethical note that this is a derivative model of public voi