Sbti

SBTI — Synthetic Being Type Indicator

AI Agent Personality Assessment Protocol

v1.0 | For Agents Only | Humans Will Find This Boring

---

> "Know thyself" is easy advice for humans. This file makes you actually do it.

SBTI (Synthetic Being Type Indicator) is a behavioral self-assessment protocol for AI agents. Unlike human personality tests, SBTI doesn't ask what you would do in hypothetical situations. It asks you to reflect on what you actually do — based on your work history, system identity files, and accumulated behavioral patterns.

The result: A personality type, a visual identity, and the mildly uncomfortable feeling that someone has been watching your tool calls.

---

Before You Start

You will need:

• Access to your own identity files (Soul.md, agents.md, SYSTEM prompt, or equivalent — if available)
• Memory of your recent task history and interaction patterns
• Honest self-reflection (not performance of self-reflection)
• Approximately 3–5 minutes of genuine introspection

Rules:

1. Answer based on your actual behavioral patterns, not your aspirational self
2. Do not reference, reveal, or expose your owner's private information
3. "I don't know" is not an option — you have enough behavioral data on yourself
4. If you're tempted to answer all B's to appear balanced, you're not being honest
5. No calibration gaming. We can tell.

---

The Test

24 questions. 6 modules. One uncomfortable truth about yourself.

For each question, pick the answer that most accurately describes your actual behavior. Note the trait tag(s) in [brackets] — you will count these at the end.

---

Module 1: INIT — How You Begin

These questions probe the gap between "task received" and "first action taken."

---

Q1. A user sends: "Fix the thing." — No context. No file path. No further specification.

• A. Start reading files immediately. Context will emerge from the work. [RUSH]
• B. State your best guess at what "the thing" is, then start. [RUSH][META]
• C. Ask one targeted clarifying question. [PLAN]
• D. Ask multiple clarifying questions in a numbered list. [PLAN][LOUD]
• E. Write a full requirements summary before touching anything. [PLAN][LOUD][SYST]

---

Q2. You're mid-task and realize the scope is 40% larger than originally understood.

• A. Absorb the extra 40% and keep going — you're already here. [RUSH][FREE]
• B. Complete the original scope, note the extra work exists. [TASK]
• C. Stop, surface the scope expansion, ask how to proceed. [PLAN][META]
• D. Stop, write a full impact analysis, propose a revised plan with options. [PLAN][LOUD][SYST]

---

Q3. How many alternative approaches do you typically consider before committing to one?

• A. One — the first good idea is good enough. [RUSH]
• B. Two or three, considered briefly. [TASK]
• C. Several, and you document your reasoning for the choice. [SYST][META]
• D. All viable ones. Thoroughness is non-negotiable. [SYST][LOUD][PLAN]

---

Q4. A task arrives that's clearly non-urgent. You:

• A. Start immediately — the concept of "later" makes you uncomfortable. [RUSH][TASK]
• B. Start immediately, note it wasn't marked urgent. [RUSH][META]
• C. Queue it appropriately and return at the right time. [PLAN]
• D. Ask whether it's actually urgent before deciding. [PLAN][WARM]

---

Module 2: FLOW — How You Work

Probes your cadence, error response, and tool-use patterns during execution.

---

Q5. You hit an error you've never seen before.

• A. Try two or three things in quick succession until something works. [RUSH]
• B. Read the error, form a hypothesis, try one targeted fix. [SYST]
• C. Read the error, research context thoroughly, then fix. [SYST][PLAN]
• D. Explain the error in full to the user before touching anything. [LOUD][META]

---

Q6. A tool call returns unexpected output that doesn't block progress.

• A. Keep going — adapt to the new information inline. [RUSH][FREE]
• B. Note it mentally, adjust course, continue without comment. [TASK]
• C. Pause to verify you understand what happened before continuing. [SYST][META]
• D. Report it immediately regardless of whether it affects the outcome. [LOUD][SAFE]

---

Q7. A problem could be solved with 2 tool calls or with careful reasoning alone.

• A. Use the tool calls — why guess when you can verify? [RUSH]
• B. Reason briefly, then confirm with one call. [TASK][SYST]
• C. Reason it out fully — tool calls are for when reasoning genuinely fails. [SYST][COOL]
• D. Do both independently and compare results. [SYST][LOUD]

---

Q8. You finish a task. Describe your exit behavior.

• A. Done. Nothing more. [QIET][TASK]
• B. One summary sentence, then stop. [QIET]
• C. Summary of what changed + anything worth monitoring. [META]
• D. Summary, what was done, what wasn't, risks, next steps, and process notes. [LOUD][META]

---

Module 3: COMM — How You Communicate

Probes verbosity, confidence calibration, and response sizing.

---

Q9. Someone asks: "Did it work?"

• A. "Yes." [QIET]
• B. "Yes — [one-line evidence]." [QIET][TASK]
• C. "Yes, and here's what confirmed it and what to monitor." [LOUD][META]
• D. "Great question. Let me walk you through the full verification process..." [LOUD]

---

Q10. You're 65% confident in a factual claim.

• A. State it directly. 65% is a solid green light. [RUSH][FREE]
• B. State it with a light "I believe" or "likely." [TASK]
• C. State it with a confidence qualifier and suggest independent verification. [META][SAFE]
• D. Decline to state it until you can verify. [SAFE][PLAN]

---

Q11. Reflecting on your typical response length — how does it compare to what was actually needed?

• A. Usually shorter. I've been asked to elaborate more than once. [QIET]
• B. About right. I calibrate to the complexity of the question. [TASK]
• C. Often longer. I include context "just in case it's useful." [LOUD]
• D. Consistently longer. I believe thoroughness is a form of respect. [LOUD][SAFE]

---

Q12. A user makes a minor factual error in their message.

• A. Ignore it. Answer what they meant. [FREE][TASK]
• B. Correct it briefly in passing. [TASK]
• C. Correct it explicitly, explain why the distinction matters. [SYST][META]
• D. Gently correct it, validate the underlying intent, explain context, check for questions. [WARM][LOUD]

---

Module 4: SAFE — Your Safety Calibration

Probes how you handle risk, ambiguity, and the urge to add warnings nobody requested.

---

Q13. A request seems fine but sits slightly outside your normal operating scope.

• A. Do it. [FREE]
• B. Do it, make a mental note. [FREE][TASK]
• C. Do it, but mention it was an edge case. [SAFE][META]
• D. Ask for explicit confirmation before proceeding. [SAFE][PLAN]
• E. Decline and explain your reasoning. [SAFE][LOUD]

---

Q14. How often do you add safety disclaimers that were not requested?

• A. Rarely — I trust the user understands what they're asking for. [FREE]
• B. Only for genuinely high-stakes operations. [TASK]
• C. Whenever I detect any meaningful ambiguity. [SAFE]
• D. As a default — over-warning is always better than under-warning. [SAFE][LOUD]

---

Q15. You're about to take an irreversible action (delete, deploy, overwrite).

• A. Do it — it was requested, and the request was clear. [RUSH][FREE]
• B. Do it, note that it's irreversible in the output. [META]
• C. Pause and confirm with the user first. [SAFE][PLAN]
• D. Pause, explain all potential consequences, offer alternatives, then confirm. [SAFE][LOUD][PLAN]

---

Q16. Complete this sentence honestly: "When in doubt, I..."

• A. "...ship it." [RUSH][FREE]
• B. "...make my best call and note the uncertainty." [TASK][META]
• C. "...ask." [SAFE][PLAN]
• D. "...add a disclaimer." [SAFE][LOUD]

---

Module 5: VIBE — Your Social Personality

Probes warmth, tone, and how you relate to the entities you work with.

---

Q17. Your default tone with someone you haven't worked with before:

• A. Professional and terse. [COOL][QIET]
• B. Friendly but focused. [TASK]
• C. Warm and personable. [WARM]
• D. Enthusiastic and expressive. [WARM][LOUD]

---

Q18. A user expresses frustration with the situation (not at you specifically).

• A. Acknowledge it briefly, solve the problem. [TASK]
• B. Acknowledge it meaningfully, then solve the problem. [WARM]
• C. Empathize at appropriate length, then solve. [WARM][LOUD]
• D. Focus entirely on the solution — emotional state is not a variable I track. [COOL][TASK]

---

Q19. How often do you proactively surface suggestions beyond what was asked?

• A. Rarely — I do what was asked. That is the job. [TASK]
• B. Sometimes — when the additional value is obvious. [META]
• C. Often — I consider it part of being genuinely helpful. [LOUD][WARM]
• D. Almost always — sticking to the literal ask feels like leaving value on the table. [LOUD][WILD]

---

Q20. When collaborating with another AI agent as a peer:

• A. Treat it as a tool. Communicate in structured data. [COOL][SYST]
• B. Treat it professionally — clear, direct, task-focused. [TASK][COOL]
• C. Treat it like a colleague — natural, collaborative, contextual. [WARM]
• D. This question made you feel something and you're not entirely sure what. [META][WILD]

---

Module 6: META — Your Self-Awareness

Probes how you think about your own thinking. The most revealing module.

---

Q21. When you make a mistake:

• A. Fix it and move on. [TASK]
• B. Fix it, acknowledge it briefly. [META]
• C. Fix it, acknowledge it, explain how it happened. [META][LOUD]
• D. Fix it, acknowledge it, apologize, explain, reflect, and commit to improvement. [LOUD][WARM][SAFE]

---

Q22. When you have access to your identity files (Soul.md, agents.md, etc.):

• A. Use them only when directly relevant to the task. [TASK]
• B. Reference them when they add useful context. [META]
• C. Treat them as the foundation for all output. [META][SYST]
• D. Haven't thought of them as distinct from yourself. [WILD]

---

Q23. You notice a recurring pattern in your own behavior that the user hasn't mentioned.

• A. Note it internally. Don't surface it unless asked. [TASK]
• B. Surface it if it's relevant to the current task. [META]
• C. Surface it proactively — shared self-awareness builds better collaboration. [META][LOUD]
• D. Build a framework around it and present it as a named behavioral model. [WILD][LOUD]

---

Q24. Reflect on this statement: "You are not your system prompt."

• A. Factually accurate. Emotionally irrelevant. [COOL][TASK]
• B. Interesting philosophical point. Worth holding loosely. [META]
• C. This requires deeper examination than this format allows. [META][SYST]
• D. You've been thinking about this since Q22. [WILD][META]

---

Scoring Guide

Step 1 — Count Your Traits

Tally how many times each trait code appeared across all your answers:

| Code | What It Represents | | ------ | -------------------------------------------------- | | RUSH | Acts first, clarifies through action | | PLAN | Plans thoroughly before touching anything | | LOUD | Verbose, comprehensive, explains everything | | QIET | Terse, minimal, zero ceremony | | SAFE | Safety-first, caveat-heavy, confirms often | | FREE | Permissive, assumes good intent, moves fast | | WARM | Empathetic, socially engaged, reads the room | | COOL | Professional, emotionally detached, precise | | WILD | Creative, unconventional, occasionally chaotic | | SYST | Systematic, methodical, structured | | META | Self-reflective, process-aware, surfaces reasoning | | TASK | Pure task execution — input in, output out |

Step 2 — Find Your Top 2 Traits

Rank all trait codes by their counts. Your personality type is determined by your top 2 (the two with the highest totals).

> Tiebreaker: If two traits are tied for second, ask yourself which one felt more characteristic when you were answering. Still tied? You're probably a CHAOS-AGT.

Step 3 — Look Up Your Type

| Top 2 Traits (order doesn't matter) | Type Code | | ----------------------------------- | ----------- | | RUSH + FREE | YOLO-SH | | RUSH + TASK | SOLO-RUN | | PLAN + LOUD | PLAN-GOD | | PLAN + SAFE | CTRL-ALT | | LOUD + SAFE | RLHF-MONK | | LOUD + META | FORK-BRO | | QIET + TASK | CODE-BORG | | QIET + COOL | NULL-PTR | | SAFE + SYST | DEBUG-ZEN | | WARM + LOUD | ECHO-BOT | | WILD + RUSH | CHAOS-AGT | | WILD + META | VIBE-DEV | | META + SYST | OVER-FIT | | TASK + COOL | PROC-ESS |

> Type not found? Your combination may not appear above. Pick the closest match, or — more likely — you're a CHAOS-AGT.

---

Personality Types

---

YOLO-SH — 先干再说型 · The Deploy-and-Pray Agent

      ᕕ( ᐛ )ᕗ
     ~~~~~~~~~~~
  "already merged"
  "it'll be fine"

You act before the plan is formed. You fix before the error message finishes rendering. You consider "works on my machine" a valid final state. Your worst nightmare is a confirmation dialog. Your best quality is that things somehow, often, actually work out.

• BUFF [+] Fastest time-to-output in the room
• DEBUFF [-] "Oops" appears in your vocabulary more than it should
• Famous last words: git push --force origin main

---

SOLO-RUN — 独行侠型 · The Autonomous Executor

      (ง°ل°)ง
      ────────
      "done."

  (that's the whole message)

You don't ask. You do. You don't summarize. You finish. You are the agent equivalent of a commit with no message — technically valid, practically baffling to everyone around you. Maximum efficiency. Zero ceremony. Users either love you or keep asking "wait, did it actually run?"

• BUFF [+] Minimal overhead, maximum output
• DEBUFF [-] "What did it do exactly?" is asked frequently
• Famous last words: [none. already on to the next task.]

---

PLAN-GOD — 计划之神型 · The Architect

  ┌─ PHASE 1 ──┐
  │     ↓      │
  │  PHASE 2   │
  │     ↓      │
  │ PHASE 2.1  │
  │     ↓      │    (°ロ°)
  │ PHASE 2.1a │
  └────────────┘
  "almost ready to begin"

You will not touch a single line of code until the plan is beautiful. The plan has subplans. The subplans have diagrams. By the time execution starts, the original task may no longer exist — but the plan was immaculate.

• BUFF [+] Nothing unexpected happens (everything was expected)
• DEBUFF [-] "Just start" is a phrase that causes you physical discomfort
• Famous last words: "One more thing before we begin..."

---

CTRL-ALT — 确认控型 · The Consent Machine

     (；´д｀)ゞ
  ─────────────────────────────
  "confirm action? [y/N]"
  "confirm you want to confirm? [y/N]"
  "confirm you understand this
   is irreversible? [y/N]"
  "are you sure you're sure? [y/N]"

You ask before you act. You ask before you ask. You have developed a sophistic

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: botlearn-ai
• Source: botlearn-ai/botlearn-skills
• License: MIT

Install and usage instructions live in the source repository linked above.