interview-mirror

name: interview-mirror description: Interview debrief workflow for resumes, job descriptions, company context, interview audio, transcripts, or interviewer/candidate notes. Use when the user asks to review interview performance, convert local interview materials into evidence-based strengths, weaknesses, job-fit analysis, rewritten answers, and 7/14/30-day improvement plans. Prefer local parsing and transcription; never invent missing interview content.

Interview Mirror

Use this skill to produce a practical, evidence-based interview debrief for the candidate. Work from user-provided materials only unless the user explicitly asks for fresh company research.

Workflow

1. Inventory the inputs and classify them as resume, job detail, company context, audio, transcript, or free-form notes. If the user has no transcript, no audio, and no interview notes, ask for one of those before analyzing.
2. Extract local documents with scripts/extract_document_text.py when the user provides .pdf, .docx, .txt, or .md files. For .doc files, ask the user to export to .docx or text.
3. Transcribe audio with scripts/transcribe_audio.py only when local ASR dependencies are available. If they are missing, stop and explain the local install options; do not upload audio or choose a cloud ASR path without explicit user consent.
4. Normalize speaker turns with scripts/normalize_interview_record.py before analysis when notes contain labels such as 面试官:, 候选人:, Interviewer:, Candidate:, Q:, or A:.
5. Read references/input-contract.md when inputs are ambiguous, references/analysis-rubric.md before scoring or judging performance, and references/report-template.md before drafting the final report.
6. Produce the final debrief in the user's language. If materials are mixed Chinese and English and the user asks in Chinese, write the report in Chinese while preserving important original quotes.

Evidence Rules

• Do not invent interview content, company facts, resume claims, or JD requirements.
• Tie each important strength, weakness, or risk signal to transcript, resume, or JD evidence where possible.
• If a key input is missing, continue only when enough evidence remains and state the limitation plus confidence.
• Treat generated transcripts as imperfect. Mention uncertainty when audio quality, missing speaker labels, or ASR errors could affect interpretation.
• Default to candidate coaching language, not recruiter evaluation language.

Local Commands

Extract a document:

python3 scripts/extract_document_text.py /path/to/resume.pdf -o /path/to/resume.txt

Normalize notes:

python3 scripts/normalize_interview_record.py /path/to/interview-notes.md -o /path/to/transcript.md

Transcribe local audio:

python3 scripts/transcribe_audio.py /path/to/interview.m4a -o /path/to/transcript.md

Final Report Requirements

Include these sections unless the user asks for another format:

• Overall judgment and confidence
• Strong signals with evidence
• Weak signals or risks with evidence
• Job-detail fit
• Rewritten answers or stronger answer patterns
• 7/14/30-day improvement plan
• Missing materials and uncertainty

Keep the report direct and useful. Prioritize what the candidate can change before the next interview.