interview-question-bank

name	interview-question-bank
description	Generate and review InterWise question-bank import packages from Codex. Use when the user asks to prepare PDF, DOCX, TXT, MD, or JSON materials for import into a private position knowledge base, or review generated knowledge atoms.

Interview Question Bank

Use this skill to prepare local JSON import packages for a user-owned position knowledge base. The application imports packages from the 岗位 / 题库维护 panel into the selected private position as draft atoms; publication and Qdrant sync are performed explicitly from that panel after review.

Workflow

1. Inspect the requested source and target role/category. Read backend/src/main/resources/application.yml's interview.position-categories to find the valid category list for the target position. Confirm which private position knowledge base the package is for.
2. Convert source files with scripts/question_bank_import.py when inputs are PDF, DOCX, TXT, MD, or JSON.
3. Review the generated package before handing it to the user. Use DRAFT for normal imports. Do not rely on package-level auto publication for user-owned knowledge bases.
4. In the web application, sign in as the owning user and open 岗位 / 题库维护.
5. Select the private position, upload the generated JSON package in 导入包维护, validate it, then import it as draft.
6. Review the imported atoms in the same panel, then publish, archive, or rebuild vector indexes explicitly.

Commands

Generate from a single-subject document (all atoms share one category):

python scripts/question_bank_import.py --input <source-dir-or-file> --category <category> --mode DRAFT

Generate from a multi-topic document (LLM picks the best category per atom from the position's valid list):

python scripts/question_bank_import.py --input <source-dir-or-file> --categories <cat1,cat2,...> --mode DRAFT

Normalize existing JSON atoms:

python scripts/question_bank_import.py --input <atoms.json> --category <category> --mode DRAFT

Category selection: The <category> or <categories> values must come from the interview.position-categories mapping for the target position in backend/src/main/resources/application.yml. Never invent category names. If the source document contains mixed topics and the target position has multiple valid categories, prefer --categories with the full valid category list for that position.

Mode Choice

• DRY_RUN: prepare a package intended for validation only.
• DRAFT: prepare atoms to be imported as drafts; they are not used by the interviewer until published later.
• AUTO_PUBLISH: legacy package mode for global maintenance only. The user-owned knowledge-base import path forces package imports to draft and requires explicit publication from the web panel.

References

• Read references/atom-schema.md before editing import package structure.
• Read references/review-rubric.md before reviewing generated atoms for quality.