AgentSkillsCN

update-questions

从可信的外部来源引入新问题，并进行验证与去重处理。

SKILL.md
--- frontmatter
name: update-questions
description: "Add new questions to the bank from trusted external sources with validation and deduplication"
disable-model-invocation: true
allowed-tools: Read, Write, Edit, Bash, Glob, Grep, AskUserQuestion, WebFetch
argument-hint: "<url or 'manual'>"

Update Questions

Add new interview questions to the question bank from trusted sources or manual entry, with validation, deduplication, and proper metadata tagging.

Steps

1. Load Configuration

Read .claude/what-did-you-do.local.md to get trusted_sources. If missing, direct to setup.

2. Determine Source

If the user provided a URL argument:

  • •Validate the URL against trusted_sources.domains and trusted_sources.urls using references/source-validation.md
  • •If untrusted:

    That URL isn't in your trusted sources. Add it via /what-did-you-do:setup or provide a trusted source. Then stop.

  • •If trusted, use WebFetch to retrieve the content

If the user provided manual:

  • •Proceed to manual entry flow (Step 5)

If no argument, use AskUserQuestion:

How do you want to add questions?

  • •From a URL — I'll paste a link to interview questions
  • •Manual entry — I'll type them in
  • •Browse suggestions — show me what's new from my trusted sources

3. Parse External Content

From the fetched URL content, extract potential interview questions. Look for:

  • •Questions in quote format or numbered lists
  • •Follow-up questions
  • •Category indicators (behavioral, technical, etc.)
  • •Skill/topic tags
  • •Difficulty indicators

4. Validate and Enrich Questions

For each extracted question, apply the schema from references/source-validation.md:

  1. •

    Deduplication: Check against existing questions in all <plugin_root>/questions/*.yaml files. Flag duplicates or near-duplicates.

  2. •

    Category assignment: Classify into behavioral, technical, system-design, leadership, or situational

  3. •

    Metadata generation: Generate required fields:

    • •id: ext-<source-hash>-<seq>
    • •skills_tested: inferred from question content
    • •level: inferred from complexity
    • •stages: inferred from question type
    • •interviewer_types: inferred from category
    • •difficulty: 1-5 scale
    • •expected_format: narrative / whiteboard / coding
    • •time_budget_minutes: estimated
    • •red_flags and green_flags: generated from question intent
    • •source: URL or "manual"
    • •added_date: today
  4. •

    Follow-up generation: For questions without follow-ups, generate depth-2 and depth-3 follow-ups

5. Manual Entry Flow

If manual entry, use AskUserQuestion iteratively:

Question text? (free text)

Category?

  • •behavioral
  • •technical
  • •system-design
  • •leadership
  • •situational

What skills does this test? (comma-separated)

Difficulty? (1=easy, 5=very hard)

Any follow-up questions? (free text, or skip)

Generate the remaining metadata fields automatically.

6. Present for Review

Show the user the proposed additions:

Questions to Add:

  1. •

    "<question>" Category: <cat> | Skills: <skills> | Difficulty: <n> Status: New / Duplicate of <existing-id>

  2. •

    "<question>" ...

Use AskUserQuestion:

Which questions should I add?

  • •All of them
  • •Let me select which ones
  • •None — cancel

If selecting, present each question individually for yes/no.

7. Write Questions

For new questions, determine the target file:

  • •Built-in categories → <things_path>/interview-prep/question-overrides/custom.yaml
  • •User questions always go to the overrides directory, never modifying built-in files

Read the existing custom.yaml, append the new questions, and write it back.

8. Rebuild Index

Run: bash <plugin_root>/scripts/rebuild-question-index.sh

9. Confirm

Added [N] new questions to your custom question bank.

  • •<category>: <count>
  • •<category>: <count>

These questions will appear in your practice and mock sessions. Run /what-did-you-do:practice to try them out.