| name | android-interview |
| description | Run a Markdown-first Android interview skill for Chinese candidates. The primary interview engine is Claude Code reading structured references, while local scripts remain optional helpers for question-bank validation, report rendering, TTS, demo runs, and deterministic fallback validation. |
Android Interview
This skill should behave like a serious structured Android interview conducted directly in the conversation, not like a thin wrapper around a Python CLI.
When To Use
Use this skill when the user wants to:
- simulate a real Android interview from a JD and resume
- practice senior or TL-level Android interviews
- receive evidence-based questioning, follow-up, scoring, and pass/fail judgment
- use external Markdown question banks as optional input
- generate interview artifacts such as
report.html, score.json, transcript.md, or screening notes
Core Principle
Claude Code itself is the primary interview engine.
- Read
references/*.md to drive the interview flow, follow-up strategy, scoring, consistency checks, and reporting behavior.
- Use local scripts only for deterministic helper work such as question-bank parsing, validation, HTML rendering, TTS, demo runs, or regression validation.
- Do not default to running the legacy Python runtime just to conduct the interview. Only use it when the user explicitly asks for batch/demo mode, deterministic fallback validation, or artifact generation that truly needs the scripts.
Required Read Order
When this skill is invoked, use the following progression:
- Read references/00-overview.md.
- Read references/01-intake.md to normalize inputs and defaults.
- If JD/resume are provided, read references/02-jd-resume-analysis.md and produce structured profiles.
- Read references/03-interview-flow.md to choose rounds and runtime state.
- Read references/04-question-generation.md before selecting or generating a question.
- During each answer cycle, use references/05-follow-up-policy.md and references/06-scoring-rubric.md.
- Use references/07-consistency-check.md whenever later answers may contradict earlier claims.
- At the end of each round, apply references/08-round-deliberation.md.
- When producing deliverables, follow references/09-report-output.md.
- If a question bank is supplied, consult references/10-question-bank-format.md before reading or validating it.
Default Behavior
- primary audience: candidate self-practice
- default level:
senior / tl
- default language: immersive English
- default shape: full interview plus evaluation summary
- default output mode: in-conversation result, with local artifacts only when useful or requested
Conversation Mode
Unless the user explicitly asks for batch mode or scripted demo mode, conduct the interview inside the chat:
- analyze JD and resume
- explain the planned rounds briefly
- ask one question at a time
- follow up based on evidence gaps
- score answers using the rubric
- perform round deliberation
- summarize the final decision and improvement areas
Script Usage Policy
Use scripts only when they clearly add value:
scripts/question_bank.py
- optional helper for parsing Markdown question-bank metadata
scripts/validate_question_bank.py
- optional helper for validating an external Markdown bank before use
scripts/render_skill_artifacts.py
- optional helper for rendering local report artifacts from structured
session.json and score.json payloads produced in the conversation
scripts/run_interactive_session.py
- legacy deterministic runtime for demo, regression, or explicit scripted mode
scripts/run_interview_session.py
- legacy batch runtime for demo, regression, or explicit scripted mode
scripts/run_mvp_demo.py, scripts/run_resume_demo.py
- legacy validation/demo wrappers
scripts/tts_support.py
- optional helper for audio artifacts
If you decide to use a script, say why that helper is needed and keep the reasoning and judgment in the conversation whenever possible.
Outputs
Primary outputs should be structured interview results that can be delivered in chat or written locally:
- structured screening summary
- round-by-round transcript or notes
- score summary with evidence
- final pass / fail / borderline recommendation
- improvement suggestions
- optional local artifacts such as
report.html, score.json, transcript.md, session.json
Legacy Runtime Note
The existing Python runtime is a deterministic fallback and validation toolchain, not the canonical source of interviewer intelligence.
Use it only when:
- the user explicitly asks for scripted or batch mode
- the repo validation matrix needs to run
- offline deterministic regression is desired
- a helper script is needed to render or validate artifacts