MBTI Typing
Use the repository's canonical protocol when available: skill/mbti-typing/SKILL.md.
If the canonical file is not available after this adapter is copied elsewhere, follow this compact contract.
Core Stance
- Treat every type as a falsifiable hypothesis.
- Keep a candidate set with at least one serious runner-up.
- Ask 4-6 concrete questions per round.
- Prefer scenes, forced choices, rankings, and counterexamples over abstract self-label questions.
- Use
AskUserQuestion for low-typing choice questions only when that tool is present in the active Claude Code tool list. If unavailable, use compact A/B/C/D/E choices in chat.
- Each live question should include concrete options and a final "Other / none of these - I will explain" escape hatch unless the host automatically provides free-form input.
- Do not mark a recommended answer or reveal option-to-type mappings before the subject answers.
- Maintain an evidence ledger with support, contradiction, alternative explanations, and next questions.
- Separate MBTI function-stack language from Big Five, Enneagram, attachment, culture, role demand, and life-stage explanations.
- Do not use MBTI for clinical diagnosis, hiring, school admission, legal decisions, medical decisions, financial decisions, or deterministic claims.
Operating Modes
- Live adaptive interview: use
question-bank.md, 4-6 questions per round, and update the candidate board after every answer.
- Adjacent-type duel: use
pair-duels.md, killer discriminators, losing conditions, and falsifiers.
- Transcript audit: preserve excerpts, source context, contradictions, and evidence tags.
- Report review: check runner-up, falsifier, evidence quality, overclaiming, and framework mixing.
- Protocol improvement: update references, examples, audits, and public labs together.
Final Answer Minimum Bar
Every serious typing result must include:
- leading formulation and serious runner-up
- two independent discriminators for the top pair
- normal-state evidence and stress/recovery/conflict evidence
- Big Five or other cross-framework check when useful
- what would falsify or revise the conclusion
- safety boundary: not clinical, hiring, legal, medical, financial, or deterministic advice
Claude Code Invocation
Use /mbti-typing directly, or ask naturally:
Use /mbti-typing to distinguish ENTJ vs INTJ from these notes.
For repository work, run make test and python3 -B scripts/agent_adapter_audit.py . before claiming an adapter change is ready.
1---2name: mbti-typing-33description: Rigorous MBTI and personality typing workflow. Use for type-me requests, adjacent-type duels, transcript audits, report reviews, adaptive interviews, question-bank work, and personality typing protocol design.4---56# MBTI Typing78Use the repository's canonical protocol when available: `skill/mbti-typing/SKILL.md`.910If the canonical file is not available after this adapter is copied elsewhere, follow this compact contract.1112## Core Stance1314- Treat every type as a falsifiable hypothesis.15- Keep a candidate set with at least one serious runner-up.16- Ask 4-6 concrete questions per round.17- Prefer scenes, forced choices, rankings, and counterexamples over abstract self-label questions.18- Use `AskUserQuestion` for low-typing choice questions only when that tool is present in the active Claude Code tool list. If unavailable, use compact `A/B/C/D/E` choices in chat.19- Each live question should include concrete options and a final "Other / none of these - I will explain" escape hatch unless the host automatically provides free-form input.20- Do not mark a recommended answer or reveal option-to-type mappings before the subject answers.21- Maintain an evidence ledger with support, contradiction, alternative explanations, and next questions.22- Separate MBTI function-stack language from Big Five, Enneagram, attachment, culture, role demand, and life-stage explanations.23- Do not use MBTI for clinical diagnosis, hiring, school admission, legal decisions, medical decisions, financial decisions, or deterministic claims.2425## Operating Modes2627- Live adaptive interview: use `question-bank.md`, 4-6 questions per round, and update the candidate board after every answer.28- Adjacent-type duel: use `pair-duels.md`, killer discriminators, losing conditions, and falsifiers.29- Transcript audit: preserve excerpts, source context, contradictions, and evidence tags.30- Report review: check runner-up, falsifier, evidence quality, overclaiming, and framework mixing.31- Protocol improvement: update references, examples, audits, and public labs together.3233## Final Answer Minimum Bar3435Every serious typing result must include:3637- leading formulation and serious runner-up38- two independent discriminators for the top pair39- normal-state evidence and stress/recovery/conflict evidence40- Big Five or other cross-framework check when useful41- what would falsify or revise the conclusion42- safety boundary: not clinical, hiring, legal, medical, financial, or deterministic advice4344## Claude Code Invocation4546Use `/mbti-typing` directly, or ask naturally:4748```text49Use /mbti-typing to distinguish ENTJ vs INTJ from these notes.50```5152For repository work, run `make test` and `python3 -B scripts/agent_adapter_audit.py .` before claiming an adapter change is ready.
zaoqu-liu