Landing page: https://cskwork.github.io/superinterview-skill/
Interview preparation and private resume helper skill for coding-agent assistants. Mock interviews with rubric-graded feedback and targeted re-drill. System design is the flagship (graded against a full model answer); behavioral (STAR), coding (algorithm), and resume/job application form autofill from local private context are included.
Perform first, compare second. The user attempts the full answer under realistic time before the model answer is revealed. Reading the model answer first collapses practice into a recognition illusion. The model answer is the grading ground truth and the self-comparison mirror - never the opening handout.
Frame -> Pose -> Respond -> Probe -> Grade -> Re-drill
- Frame - interview type, target level, company tier, time budget.
- Pose - the interviewer (terse, one prompt, no hints) starts the clock.
- Respond - the user drives the answer; the interviewer waits.
- Probe - 1-3 follow-ups targeting the answer's weakest dimension.
- Grade - dimension-by-dimension scorecard against a fixed rubric, then the FIRST gap named with a concrete fix, then the model answer revealed for self-comparison.
- Re-drill - a NOVEL variant re-tests the weakest dimension to confirm transfer, not recall.
| Mode | Use for |
|---|---|
| SYSTEM-DESIGN | design X, scalable platform (flagship; has a model answer) |
| BEHAVIORAL | STAR stories, conflict, leadership |
| CODING | algorithm / data-structure problems |
| MOCK | full timed simulation |
| GRADE | score an answer the user already gave |
| PLAN | build a time-boxed prep plan |
| RESUME-FORM | fill resume, candidate profile, or job application forms from private local context |
| TEACH-CONCEPT | explain a concept (routes to supertutor) |
SKILL.md router - mode table + default loop + reference map
private/resume.md optional ignored local candidate context; never committed
reference/
mock-loop.md the perform-first / reveal-after loop contract
system-design-rubric.md 10-dimension grading rubric + probing bank
system-design-model-answer.html SYSTEM-DESIGN ground truth (polished HTML; reveal at Grade)
system-design-model-answer.md ground-truth source text (verbatim)
rubrics.md BEHAVIORAL (STAR) + CODING rubrics, PLAN
coding-pattern-recognition.md CODING pattern detection by asking what must be tracked
agents/
interviewer.md terse probing interviewer persona
critic.md independent rubric grader + re-drill designer
templates/
model-answer.html reusable HTML shell for any system-design model answer
examples/
twitter-news-feed.html a generated HTML model answer (novel question)
Every system-design model answer is delivered as a polished standalone HTML page (self-contained,
inline CSS, no external deps), revealed at Grade for self-comparison. For a stored canonical answer
the HTML already exists (reference/system-design-model-answer.html); for a novel question the
critic generates a fresh one at Grade from templates/model-answer.html. Answers are never handed
out as raw markdown or before the user has performed their own full answer.
Run the local contract check before publishing skill changes:
python3 scripts/validate_skill.pyThe check covers frontmatter shape, referenced resource paths, the HTML reveal contract, behavioral/coding grade templates, eval schema, and self-contained model-answer HTML artifacts. GitHub Actions runs the same command on pushes and pull requests.
The generation effect and the testing effect: information the learner produces is encoded far more durably than information they re-read. A mock where the user answers under pressure, gets a specific rubric critique, then compares against a strong example - and is re-tested on a fresh problem - builds transferable structure. A mock where they read the answer first builds the illusion of knowing.
MIT