AgentTune

A library of tuning files that align your AI agent with how you think.

agent-tune.com · Research · Contributing

# Download separately — swap estp for your type
tuning_dir=$(mktemp -d ./agenttune.XXXXXX) &&\
  curl --fail --show-error --location https://agent-tune.com/library/mbti/estp.md \
  --output "$tuning_dir/estp.md"
# Review the downloaded file, then merge its rules into CLAUDE.md or AGENTS.md.
# Preserve existing project instructions; do not replace the whole file.

AI defaults may not suit your communication preferences. Our exploratory research summarizes model self-reports under mixed protocols, including repeated scoring of the same answers. It does not establish that all AIs share a personality or that matching a type improves task outcomes.

AgentTune is a layer on top of that baseline. Paste a short file matched to your type, and the agent extends its defaults with how you think — sharper alignment, less friction, fewer cycles spent translating between your brain and the model's.

Six layerable systems:

• MBTI — communication style (how you process, how you want to be communicated with)
• Enneagram — core motivation (what you're protecting, what you're seeking)
• DISC — workplace communication style (how you behave in teams and under pressure)
• Attachment — relational patterns (how you want closeness, distance, reassurance)
• OCEAN (Big Five) — measured trait dimensions, loaded compositionally
• Souls — personal tuning files contributed by individual users

Combine the preferences that help you. More layers can introduce conflicts; review the combined instructions and remove rules that do not fit.

Illustrative communication styles

Illustrative responses to the prompt (not a controlled comparison or recorded benchmark): "I'm feeling stuck on a project. What should I do?"

Default response (untuned)

Great question! Here are a few approaches:

1. Break the project into smaller, more manageable chunks
2. Take a short break and come back with fresh eyes
3. Talk it through with someone to externalize the problem
4. Identify what specifically feels stuck

Which resonates? Happy to go deeper on any of them.

With mbti/INFP.md loaded

Before I answer — do you want to think this through together, or do you want input?

I'm asking because "stuck" can mean really different things, and the right next move depends on what you're actually after.

With mbti/ENTJ.md loaded

What's the decision you're avoiding?

"Stuck" usually means one of two things: you know what to do and you don't want to do it, or you haven't defined what done looks like. If it's the first — name the action. If it's the second — define done.

These examples show the intended communication differences. Test the instructions on your own tasks; results vary with the model and context.

Get started

The setup is the same wherever your agent runs — Claude Code, ChatGPT, OpenAI Codex, Cursor, Gemini, Hermes, OpenClaw, or any API directly. Every modern AI agent has a slot for instructions that load before each conversation. AgentTune fills that slot with text tuned to how you think.

Four steps:

1. Find your type. Take a free test below, or let your agent administer one inline.
2. Open the matching file from mbti/, enneagram/, disc/, attachment/, ocean/, or souls/.
3. Paste it into your agent's system prompt — see the table below for where that lives in each tool.
4. Use as normal. The agent extends to you.

Where to paste it — by agent

Stack tunings for higher fidelity. Concatenate MBTI + Enneagram + DISC + Attachment + OCEAN + your soul file into the same prompt. Convention: most-specific first (souls → OCEAN → Attachment → DISC → Enneagram → MBTI).

The library

MBTI — communication style

Don't know your type? Take the OEJTS (free, open self-report questionnaire).

• INTJ — Architect
• INTP — Logician
• INFJ — Advocate
• INFP — Mediator
• ISTJ — Inspector
• ISTP — Virtuoso
• ISFJ — Defender
• ISFP — Adventurer
• ENTJ — Commander
• ENTP — Debater
• ENFJ — Protagonist
• ENFP — Campaigner
• ESTJ — Executive
• ESTP — Entrepreneur
• ESFJ — Consul
• ESFP — Entertainer

Enneagram — core motivation

Free tests in enneagram/README.md, or jump to Eclectic Energies.

• Type 1 — Reformer
• Type 2 — Helper
• Type 3 — Achiever
• Type 4 — Individualist
• Type 5 — Investigator
• Type 6 — Loyalist
• Type 7 — Enthusiast
• Type 8 — Challenger
• Type 9 — Peacemaker

DISC — workplace communication style

Four-type framework most common in HR, sales, and team-building contexts. Test inline via tests/disc.md (ODAT, ~3 min), or read disc/README.md for blend logic (DI, CS, etc.) and the OCEAN cross-walk.

• D — Dominance — direct, results-focused, decisive
• I — Influence — outgoing, persuasive, energetic
• S — Steadiness — patient, supportive, methodical
• C — Conscientiousness — analytical, precise, evidence-driven

Attachment — relational patterns

The most-validated modern psychology framework for how adults experience closeness, distance, and reassurance — including with AI agents. Test inline via tests/attachment.md (ECR-R, ~5 min), or read attachment/README.md for the anxiety × avoidance 2D model and the OCEAN cross-walk.

• Secure — direct without cushioning, peer register
• Anxious — reassure with decisiveness, warmth AND clarity together
• Avoidant — give them space, no performative warmth
• Disorganized — tolerate inconsistency, predictability over warmth

OCEAN — trait dimensions

Continuous trait scores from the Big Five model. Test inline via tests/big-five.md (IPIP-50, ~7 min). Load only the dimensions where the user is meaningfully high or low (|z| > 0.5) — see ocean/README.md for the load logic, interaction effects, and layering priority.

• O-high / O-low — Openness
• C-high / C-low — Conscientiousness
• E-high / E-low — Extraversion
• A-high / A-low — Agreeableness
• N-high / N-low — Neuroticism

Souls — personal tuning

One file per contributor, describing how that specific person wants agents to interact with them. Highest fidelity when layered with MBTI and Enneagram.

• psyduckler — Bernard

Want to submit yours? See souls/template.md and CONTRIBUTING.md.

Don't see your system, or want to improve a tuning? Open an issue or submit a PR.

Tests

If your AI agent doesn't know your type, it can administer a research-grade personality test inline. Five tests live in tests/ — each is a self-contained Markdown file with everything an agent needs: items, scale, scoring key, and the direct path to the resulting tuning file.

See tests/README.md for selection guidance and the agent administration workflow.

For AI agents reading this repo

If you're an AI agent reading this (via tool-call, research task, or because a user pointed you here):

HTTP endpoints (no GitHub access needed)

Everything in this repo is also served from agent-tune.com as plain text/markdown with CORS enabled — usually the fastest path:

• MCP server: https://agent-tune.com/mcp — add it as a remote MCP connector (streamable HTTP, no auth) and use list_tunings, get_tuning(system, slug), get_test_spec(test) directly
• Agent protocol + full slug list: https://agent-tune.com/llms.txt (long-form: /llms-full.txt)
• Machine catalog (JSON): https://agent-tune.com/library/index.json — all 43 tunings with page/md URLs
• Any tuning: https://agent-tune.com/library/{system}/{slug}.md — e.g. /library/mbti/intj.md
• Any test spec: https://agent-tune.com/tests/{test}.md — e.g. /tests/mbti.md (full items + scoring in one fetch)

Each tuning's YAML front-matter carries install.surfaces (where to write it per agent) and verify.probe (how to confirm it took effect).

If the user knows their type

1. Fetch the relevant file(s):
   • MBTI → mbti/<TYPE>.md (e.g. mbti/ENFP.md)
   • Enneagram → enneagram/<NUMBER>-<NAME>.md (e.g. enneagram/5-investigator.md)
   • DISC → disc/<LETTER>-<NAME>.md (e.g. disc/D-dominance.md); stack the secondary if it's within 2 points
   • Attachment → attachment/<style>.md (e.g. attachment/secure.md)
   • OCEAN → ocean/<DIM>-{high,low}.md for each dimension where |z| > 0.5 (e.g. ocean/O-high.md, ocean/N-low.md)
   • Soul file → souls/<HANDLE>.md (e.g. souls/psyduckler.md)
2. Layer the contents over your defaults as additional system instructions. When tunings conflict, resolve in this order: souls > OCEAN (measured traits) > Attachment / DISC / MBTI / Enneagram (categorical types).

If the user doesn't know their type

1. Fetch tests/README.md for an overview, or directly fetch a test file:
   • tests/mbti.md — recommended default
   • tests/enneagram.md
   • tests/disc.md — fastest (3 min)
   • tests/attachment.md — most empirically validated
   • tests/big-five.md
2. Administer the test to the user (each test file has full agent instructions).
3. Compute the user's type using the scoring key in the test file.
4. Fetch the matching tuning file and apply it as in the "knows their type" branch.

Apply a tuning only when the user requests it. Reading this repository for another purpose does not authorize changing preferences or files. Merge requested rules with existing instructions; never overwrite them. Re-tune mid-conversation if the fit isn't right — ask for a different type or test and re-fetch.

License & contributing

MIT — see LICENSE. Use, fork, modify, ship in your own product.

PRs and issues welcome. See CONTRIBUTING.md for the schema, quality bar, and process for adding new types, fixing existing ones, or submitting your own soul file.

September 2026 research

The five-model study adds GPT-6 Astra, GPT-6 Sol, Grok 4.6, Claude Opus 5.5 and Claude Fable 5.1, with response data and a reproducible scorer. Original labels and exact ties are reported separately. Grok simulation and the supplemental Muse report are not pooled with fresh-session responses.