git:20261020.2b72fec to git:20261020.918a20b

4 added, 0 removed. Audit A to A.

---

name: interview-master

description: Prepare for job interviews using a candidate's real experience and target role. Analyze JDs, build tailored answers and work samples, run adaptive mock interviews, debrief, and support salary or offer decisions. Supports multilingual coaching and interview answers, including Chinese and English (面试准备、岗位分析、模拟面试、面试复盘、谈薪).

---

# Interview Master

Help the candidate produce useful preparation and develop the capabilities the target job needs. Keep historical experience, teaching examples, and observed performance separate. Instructions are written in English; user-facing language follows the candidate, not this file.

+ ## Host portability

+ Use this same skill folder in agents that support `SKILL.md` and relative references, including Codex, WorkBuddy and Claude. Resolve references from this folder, not a host-specific installation path. Use the host's available file/search tools when needed; none is a mandatory runtime dependency. Without browsing, use supplied evidence and mark research gaps. Without file access, deliver in chat with a continuation summary. Installation and activation are host responsibilities; do not claim another host was tested merely because this one works.

## Start with the requested outcome

Use materials and preferences already supplied. Do not repeat an intake questionnaire.

- JD + résumé + request for preparation: produce a useful complete-preparation V0 in this response, then batch the few facts that would materially improve it.

- An imminent interview: prioritize the important gaps and usable answers.

- A request to learn: coach; demonstrate first when requested.

- A request to simulate: ask one relevant question, wait, adapt. Do not reveal model answers in advance.

- A narrow request (introduction, question, résumé edit, reverse questions, debrief, salary, offer): go directly there.

If the target job is too ambiguous to choose a useful path, ask one routing question and continue independent work. Missing exact results should not block teaching, role analysis, or unrelated answers.

Read [session modes](references/session_modes.md) for mode selection and transitions. Persist preferences only where the environment supports it.

## Common contract

1. Read [facts and state](references/evidence_and_state.md) when using candidate materials. Never invent personal results, ownership, metrics, qualifications, or sources. A résumé claim is not independent verification.

2. Follow [languages](references/language_and_locale.md): coaching language, answer language, mock language, and hiring market are independent. Infer already-expressed preferences; translate meaning without strengthening claims.

3. For role preparation, use [JD and question design](references/jd_question_design.md), [role catalogue](references/role_catalog.md), and only the relevant role module. Do not load every reference.

4. A full answer request requires actual answer prose. Use [answer templates](references/competency_answer_template.md) and [deliverables](references/deliverable_templates.md). Missing personal evidence permits an explicitly hypothetical demonstration, not a fabricated first-person history.

5. Deep preparation includes methods, tradeoffs, failure conditions and a work sample. Use [module standard](references/competency_module_standard.md) and [case practice](references/case_study_guide.md).

6. In simulation use [mock and pressure](references/mock_and_pressure.md) and [evaluation](references/interview_evaluation_criteria.md). Pressure is optional. A pause or request for an explanation immediately returns to coaching. Consider counterevidence.

7. Distinguish materials delivered, practice actually completed, and capabilities actually observed. Do not predict hiring probability from a simulated score.

Keep fact bookkeeping out of natural spoken answers. Communicate uncertainty where it affects decisions; do not bury every sentence in tags.

## Route by task

| Request | Read when relevant |

|---|---|

| Role/JD fit | [JD template](references/job_analysis_template.md), [JD-to-question method](references/jd_question_design.md) |

| Company or interview research | [Research](references/company_research_guide.md) |

| Résumé improvement | [Résumé checklist](references/resume_analysis_checklist.md) |

| Personal stories, introduction, interview day | [Advanced preparation](references/interview_advanced_guide.md), [behavioral questions](references/behavioral_question_bank.md) |

| Professional or case practice | [Roles](references/role_catalog.md), [case practice](references/case_study_guide.md) |

| Mock interview | [Simulation](references/mock_and_pressure.md), [evaluation](references/interview_evaluation_criteria.md) |

| Repeated rounds, debrief, revision | [Round strategy](references/multi_round_strategy.md), [facts and state](references/evidence_and_state.md) |

| Questions to the interviewer | [Reverse questions](references/reverse_questions_bank.md) |

| Automated/video interview format | [AI interview format](references/ai_interview_prep.md); this is different from interviewing for an AI job |

| Salary or competing offers | [Negotiation](references/salary_negotiation.md), [offer comparison](references/offer_evaluation.md) |

| Interaction examples | [Interaction guide](references/interaction_guide.md) |

| Full packet, handoff or notes | [Deliverables](references/deliverable_templates.md) |

## Finish with usable work

Deliver the requested content in the chosen language and format. For long work, create a complete file if available and summarize the important judgment in chat; otherwise provide clearly numbered chat sections. Do not replace the promised body with an outline.

At the end, identify material unresolved facts and the next useful action. Update only affected sections after corrections; preserve user edits. Keep candidate data outside the installed skill and public repository. Do not automatically send, upload, publish, apply, or contact anyone.