Pick how you like to work
Want direct answers first, clear trade-offs, honest pushback, or a warmer tone? Choose your preferences directly, or use one of the 5 free tests and 43 templates as a starting point.
Give ChatGPT, Claude, Meta Muse Soul.md, Grok, Gemini, or your coding agent (CLAUDE.md & AGENTS.md) clear rules for how to talk to you. Pick plain-English preferences, copy one of 43 MIT templates, or take a 5-minute no-email test.
Jump to setup: Meta Muse Soul.md · Claude · ChatGPT · Claude Code · Codex AGENTS.md · Grok · Gemini · MCP Server
Want direct answers first, clear trade-offs, honest pushback, or a warmer tone? Choose your preferences directly, or use one of the 5 free tests and 43 templates as a starting point.
Pick where you will paste it: ChatGPT Custom Instructions, Claude Profile Preferences, Grok, Meta Muse Soul.md, or CLAUDE.md and AGENTS.md. Edit any line before copying.
Paste the block alongside your existing instructions and run a real task. Keep the rules that help and delete any line that gets in the way.
Telling an AI to “be more direct” is too vague. Instead, tell it where to start the answer, how much detail to include, and when to push back.
For example: ask it to lead with the recommendation, name the main trade-offs, and flag anything uncertain. Your live prompt in chat still overrides these defaults whenever you need something different.
Customize in the generator → · Use the 6-choice preferences builder →
Lead with your recommendation. Explain the main reason and practical tradeoffs. Use a short list when it makes the answer clearer. State important uncertainty and relevant risks. Challenge a weak assumption with evidence. Follow my current request when it differs.
That's completely normal! Everyone hits a wall sometimes. Here are a few approaches you might try: 1. Break the project into smaller, manageable steps 2. Take a short break and come back with fresh eyes 3. Talk it through with a colleague 4. Identify what specifically feels stuck Which of these resonates most? Happy to explore any of them!
List the project's next three deliverables and mark which one has an unresolved input or decision. Being stuck is usually an unmapped dependency or an unmade trade-off rather than a lack of effort. Reply with the deliverable you are avoiding and what makes it ambiguous; we will isolate the blocker and pick one next action.
Not sure which communication rules you prefer? Take any of the five open self-report questionnaires to explore matching templates, or compare all five instruments on the Tests hub.
Use the same preference block across chat apps and coding agents, formatted for each platform's character limits and file conventions.
Platform guides: ChatGPT settings · Claude · Meta Muse · Perplexity Spaces · Coding rules matrix · Gemini · Grok · 5-System Matcher
Free browser utilities: Preferences builder · Instruction conflict checker · Format converter · Muse Soul.md editor · A/B test bench · Memory export reviewer · All 10 free tools →
We scored 2,000+ fresh AI test sessions across Claude Opus 5.5, Claude Fable 5.1, GPT-6 Astra, and GPT-6 Sol, plus a 60-reply before/after tuning pilot. These studies don't prove AI has a human personality—they show how models answer human surveys by default and how explicit rules change reply structure. Every study includes raw answers, scoring code, and limits.
Explore all research studies → · Five-model benchmark (2,000+ runs) → · Before/after tuning pilot → · Templates vs. plain-English rules →
Your quiz answers and instruction drafts stay in your browser tab and can be cleared anytime. All 43 templates and code are MIT-licensed. AI agents can read clean Markdown, local scoring specs, and safe merge rules via /llms.txt or the zero-auth MCP server at https://agent-tune.com/mcp.