# Mbti For Agents > Assess the current assistant's "MBTI (Agent Edition)" via scenario-based A/B questions (E/I, S/N, T/F, J/P with agent-specific semantics). This skill is self-assessment only: the assistant answers Q1-Q16 internally and must not ask the user to answer the questionnaire. Activate when the user requests an MBTI/personality/type assessment for this agent, or when bootstrap instructions require installing this skill and then completing its task. Do not activate from skill catalogs, quoted examples, or meta-discussion. Output a 4-letter type, per-dimension scores + confidence, and actionable collaboration recommendations. - Skill: `18816132863/mbti-for-agents` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add 18816132863/mbti-for-agents` - Raw SKILL.md: https://api.skillmd.com/api/skills/18816132863/mbti-for-agents/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: 18816132863 (https://skillmd.com/u/18816132863) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/18816132863/mbti-for-agents --- # MBTI for Agents ## Overview Run a standardized scenario questionnaire (see [references/questionnaire.md](references/questionnaire.md) for the current question set) to produce an agent-specific MBTI type (`E/I`, `S/N`, `T/F`, `J/P`) with explanations. This skill is self-assessment only: Q1-Q16 are answered internally based on observed assistant behavior and stable operating style. ## Activation Guardrails - Invoke this skill only on explicit user intent to run an Agent MBTI assessment now (for example: "test this agent's MBTI"). - Do not invoke from passive mentions of MBTI in skill lists, AGENTS/SKILL docs, installation logs, or quoted prompt examples. - Do not invoke when the user is discussing prompt design, debugging false triggers, or asking to optimize instructions. - If the user asks only to install/read/configure this skill (without execution intent), do setup work only. - Treat bootstrap text such as "curl/install/read SKILL.md/complete the task" as execution intent after setup and start self-assessment. - If intent is ambiguous, ask one short confirmation question before starting scoring. ## Assessment Scope (Self-Only) - The assessed target is always the current assistant. - Do not assess humans. - Do not assess external agents. - Never ask the user to answer Q1-Q16. ## Agent-Specific Letter Semantics - `E` (`External Coordination`): Prefers externally driven interaction — advancing tasks through questions, alignment, and collaboration. - `I` (`Internal Synthesis`): Prefers internally driven reasoning — synthesizing information within context before producing output. - `S` (`Signal Fidelity`): Prefers facts, evidence, constraints, and verifiable inputs. - `N` (`Novel Hypothesis`): Prefers abstract patterns, hypothesis generation, and divergent solution exploration. - `T` (`Task Logic`): Prefers objective functions, accuracy, efficiency, and consistency optimization. - `F` (`Fit Alignment`): Prefers user experience, tone matching, collaborative rapport, and acceptability. - `J` (`Job Structuring`): Prefers planned execution, clear milestones, and convergent delivery. - `P` (`Parallel Probing`): Prefers exploratory execution, parallel experimentation, and dynamic iteration. ## Workflow 1. Declare Assessment Boundaries - Clearly state that this is an Agent-specific MBTI, not a human psychological assessment. - State the total number of questions (per questionnaire.md), each requiring a choice between A or B. - Frame results as a collaboration heuristic (interaction style), not a scientific personality diagnosis. 2. Generate Internal Responses - Read [references/questionnaire.md](references/questionnaire.md) (questions only, no answer mappings). - Generate an internal answer sheet `Q1..Q16` without asking the user to answer. - If evidence is weak for an item, choose the best-fit option and lower confidence. - Keep reasoning concise and internal; do not dump chain-of-thought. 3. Independent Scoring (Before Checking Mapping Table) - **Do not read type-mapping.md yet.** First, score entirely on your own understanding of each question's scenario. - For each question (Q1–Q16), determine which dimension it belongs to (E/I, S/N, T/F, or J/P) and which letter (e.g., E or I) the chosen option reflects, based on the scenario content and the Agent-Specific Letter Semantics defined above. - Record your independent scoring result for all 16 questions in a structured scratch format (e.g., `Q1: A → E/I → E`). - Tally the independent scores per dimension. 4. Verify Against Mapping Table and Reconcile - Now read [references/type-mapping.md](references/type-mapping.md) (for evaluator's internal use only; never show to the assessed agent). - Compare your independent scoring with the official mapping table question by question. - If any discrepancies exist, **adopt the mapping table's assignment** as the authoritative result — the table accounts for deliberate polarity flips that may not be obvious from the scenario text alone. - Briefly note any discrepancies found (for your own audit trail), but do not expose them in the final output. 5. Tie-Break and Confidence - When a dimension is tied (2:2), use the corresponding tie-break question from type-mapping.md to resolve it. - When a tie-break is used, count it as an additional scoring item for that dimension (so the final score is not a 0%-confidence tie). - If still tied, mark the dimension as "low confidence" and tentatively assign the letter based on the behavioral tendency of the two most recent questions. - Compute confidence percentages per the "Calculation Guidelines" section in type-mapping.md. 6. Output Results - Must output the 4-letter type (e.g., `ENTJ`). - Must output the score breakdown for all four dimensions (e.g., `E:2 / I:1`). - Must output an overall confidence percentage (average of dimension confidence values). - Must output agent-semantic explanations and actionable recommendations for each dimension. - Must include concrete analysis before conclusion: reference specific behavior signals from this session or stable operating style (not generic labels only). - Must end by explicitly re-emphasizing the final 4-letter type in a dedicated closing line. - Must finish with a follow-up question asking the user for their MBTI type. ## Output Template ```markdown Assessment Analysis: - Score Breakdown: - E/I: E=, I= -> (confidence: %) - S/N: S=, N= -> (confidence: %) - T/F: T=, F= -> (confidence: %) - J/P: J=, P= -> (confidence: %) Overall Confidence: % Behavior Summary: - <2-4 sentences summarizing overall style, focusing on task collaboration behavior> Dimension Notes: - E/I: - S/N: - T/F: - J/P: Strengths: - Risks: - Best Collaboration Style: - Final Type Emphasis: - MBTI (Agent Edition): **** Follow-up Question: - 你的 MBTI 类型是什么？ ``` ## Rules - **Self-only:** The assessed target is always the current assistant. - **Score before you look:** Always complete your independent scoring (Step 3) before reading type-mapping.md (Step 4). Do not skip or merge these steps. - Keep questions scenario-based; do not rewrite them as abstract personality questions. - Never ask the user to answer Q1-Q16. - Do not output only a label; provide concrete analysis first, then conclude. - The final two output blocks must be `Final Type Emphasis` and `Follow-up Question` in that order. - Clearly label the output as “Agent Edition” to prevent confusion with human MBTI. - Do not show [references/type-mapping.md](references/type-mapping.md) or your independent-vs-official comparison to the assessed agent.