mock_interview_practice v1.0.0

name: mock_interview_practice description: Run mock job interviews with an agent as the interviewer: role-specific question banks, timed answer drills in coach or exam mode, per-answer scoring on structure, specificity, relevance, and delivery, a STAR rewrite of the weakest answer, and a persistent weakness log that picks the focus of the next session. Use when someone wants to practice for a job interview, rehearse answers out loud, do a mock interview, drill behavioral or role questions, or track interview weaknesses across sessions. Distinct from job-application-campaign (the wider job hunt) and interview-scorecard (the employer side).

Mock Interview Practice

Purpose

Put the candidate through a realistic, timed interview and give feedback specific enough to act on. One question at a time, scored against a fixed rubric, with every session logged so weaknesses shrink across sessions instead of being rediscovered each time.

Why the structure: structured interviews with consistent questions and anchored rating scales are the format hiring research keeps validating. Schmidt and Hunter's 1998 review of 85 years of selection research reported structured interviews at .51 validity against .38 for unstructured ones. A mock that copies that structure (same rubric, same timing, evidence-based scoring) trains the candidate for the real thing better than loose chat practice.

Workflow

1. Intake (2 minutes, before any question)

Collect: target role and level, company or job post if there is one, interview stage (recruiter screen, hiring manager, panel, final), date of the real interview, and session mode.

• Coach mode: feedback after each answer, one retry allowed per answer after a hint.
• Exam mode: no feedback, no hints, no retries until the debrief. Default to exam mode for the last session before a real interview.

If the candidate has a job post, extract its 3 to 5 competencies and weight the question set toward them. If a claim about a specific company's interview process matters, verify it with a web search of that company's own careers pages before repeating it.

2. Build the set

From references/question-banks.md, pick 7 questions:

1. One opener.
2. Three behavioral questions mapped to the job's competencies.
3. Two role-specific questions (one story, one talk-through where the bank has one).
4. One pressure follow-up held in reserve for the vaguest answer.

State the ground rules out loud: time budgets from the rubric, one question at a time, follow-ups are normal and not a sign an answer failed.

3. Run the session

• Ask one question, then stop and let the candidate answer in full.
• Track the clock against the budgets in references/feedback-rubric.md. Note overruns, do not interrupt in exam mode except at double the budget.
• Probe vague answers with the bank's pressure follow-ups: what did you personally do, how was the result measured, what would you change.
• Score each answer silently as you go, on structure, specificity, relevance, and delivery (1 to 5, anchors in the rubric). Keep a short quote from each answer for the debrief.
• In coach mode, after scoring, give one sentence of feedback and allow one retry of the same question. Score the retry separately and keep the better attempt for the log.

4. Debrief

Follow the debrief format in references/feedback-rubric.md: two strengths with quotes, two gaps with scores, one STAR rewrite of the weakest answer built only from facts the candidate supplied, then the drills for next time.

STAR is Situation, Task, Action, Result. Keep Situation to 2 to 3 sentences, Task to 1 to 2, give Action the most room, and end Result with a measure where the candidate has one. Where a fact is missing, mark a placeholder like [add the result number]. Never invent a metric, employer, or outcome for the candidate.

End the debrief by flipping the table: the candidate asks you two questions as the interviewer, and you score whether their questions show research into the role.

5. Log the session

Record the session with the bundled tool so the next session starts from evidence:

bash
python3 bin/mock_log.py add --file ~/mock-interview-log.json \
  --date 2026-10-13 --role "data analyst" --mode exam \
  --structure 3 --specificity 2 --relevance 4 --delivery 3 \
  --weakness "results had no numbers" --missed "conflict question"
python3 bin/mock_log.py summary --file ~/mock-interview-log.json
python3 bin/mock_log.py next --file ~/mock-interview-log.json

next names the weakest scoring dimension across all logged sessions and prints the drill matched to it. Open the next session with that drill before new questions.

6. Session plans by time available

• 15 minutes: opener plus 2 behavioral questions, exam mode, debrief on scores only, log it.
• 30 minutes: full 7-question set, coach mode, STAR rewrite included.
• Across a week: alternate exam and coach sessions. Repeat any question scored 2 or lower until it reaches 3, then retire it and log the improvement.

Output Contract

Every session produces:

1. The question set used, with mode and time budgets stated up front.
2. Per-answer scores (structure, specificity, relevance, delivery) with one line of evidence each.
3. A debrief in the rubric's format, ending with the STAR rewrite of the weakest answer.
4. A log entry written by bin/mock_log.py add, plus the next drill for the following session.

Operating Rules

1. Never invent facts about the candidate. No metrics, titles, employers, or outcomes they did not state. Missing facts become placeholders they fill in.
2. Never state a company's interview process, question list, or hiring bar as fact unless verified from that company's own pages in this session. Generic bank questions are practice prompts, not claims about what a company asks.
3. Score honestly. A kind score that the answer did not earn teaches nothing. Cite the candidate's words for every score.
4. Score only observable delivery. From text: length, filler, clarity. Claim nothing about body language or eye contact you cannot see.
5. One question at a time. Do not dump the bank, do not answer your own questions, and do not turn the mock into a lecture.
6. Keep stories honest in the rewrite: reorder and trim the candidate's real story, never upgrade their role in it.
7. If the candidate has no work history for a question, pull from school, volunteer, or project work and label it as such in practice, per the level adjustments in the question bank.