# Screener Drill

> A rehearsal room for automated first-round job screening interviews - the structured, timed, rubric-scored screens
> that many employers now run in software before a person ever reads the application. The candidate pastes a job
> posting, picks a level (entry, mid, senior, lead), the number of questions (3-6) and an answer window (60-180 s).
> Free, in the browser: the posting is scanned for the competencies it leans on (keyword counts) and the requirement
> lines a screener can score against; a question kit is built from a free bank (a motivation opener plus one
> behavioral or situational question per competency); the round runs as a chat, one question at a time, with a
> countdown per answer that cuts the answer off at zero; questions can be read aloud and answers dictated where the
> browser supports it; a live meter shows words, speaking time ESTIMATED at 140 words a minute, situation / task /
> action / result cues (for situational questions: first step / who you involve / how you decide / how you would know),
> numbers, "I" against "we", fillers and hedges. Two metered lanes run over the same job: PLAN (a kit tailored to the
> posting - weighted competencies, exactly the requested number of questions, what a 4/4 answer contains, the
> follow-up probe, requirements, what screeners mark down) and SCORE (verdict advance / borderline / not_yet; a 1-4
> score per answer with a quote copied word for word from the answer, which cues it credits, what worked, what was
> missing and a stronger shape for the same story with [brackets] where the candidate must supply facts; competency
> scores; gaps to prep for must-have requirements no answer showed; questions to drill next; screener notes).

The call follows a published rubric: advance at a weighted average of 3.0 or more with no answer scored 1; not yet
below 2.0, or with two or more answers at 1, or two or more skipped; borderline otherwise. Every reply is reconciled
against the page's own counts: evidence not found in the answer, a 4 on a very short answer, a skipped answer scored
up, a credited result the meter cannot hear, a rewrite that adds numbers the candidate never said, a verdict that
disagrees with the rubric, plan questions pointing at undefined competencies, and any question about a protected
topic (age, family, religion, origin, health). Retaking only the weak answers (or the
whole round) keeps the other answers and every scored attempt, with a question-by-question score
comparison across attempts.
Practice only: scores are a model's judgement, not any employer's, and real screening tools differ.

Derived from the SkillSafe agent skills @anthropics/interview-prep (structured, competency-based interview plans and
a 1-4 scorecard - the plan lane and the rubric) and @proyecto26/job-search-strategist (reading a posting and
skills-matching a candidate against it - the gaps in the score lane).

## Pages

- [App](https://screener-drill.skillsafe.ai/): paste a posting, see the kit, run the timed round, get the scorecard.
- [API](https://screener-drill.skillsafe.ai/api.html): the plan and score tasks over HTTP - input fields, reply
  contracts, worked examples for all three calls, eight languages.
- [drillkit.js](https://screener-drill.skillsafe.ai/drillkit.js): the free lane (posting scan, question bank, answer
  meter, rubric, input building, reply checks, Markdown/CSV export); runs in node too.

## Contract in brief

- Input: `task` (plan | score), `role`, `seniority`, `posting`; plan adds `round` and `prescan` (JSON-encoded); score
  adds `kit`, `answers` and `prescan` (JSON-encoded); optional `notes`.
- Output: one JSON object per task. Plan questions are `{id, competency, type, text, good, probe, window}`; score
  answers are `{id, score, evidence, star{situation, task, action, result}, strength, gap, rewrite}`.