interview-panel
/interview-panel
Rehearse a panel job interview by roleplay: two to four interviewers (Head of Cloud, Head of HR, Head of Engineering, Lead Engineer, a Principal running system-design scenarios, a bar raiser) ask one question at a time and follow up in character, then an independent reviewer debriefs the transcript against the candidate's evidence base. Use for a mock interview, interview practice or panel prep for Lead, Staff or Principal cloud, platform or software roles. Not for a CV or cover letter.
- Author
- kevin-burns ↗
- Repository
- claude-skills ↗
- Licence
- MIT
- Last repository change
- 21.10.2026
Assessment and security
- Files in this package
- 18
- Scripts
- 2
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 21.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- 2 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: scripts/check_bank.py, tests/test_check_bank.py
What it does and when to use it
Rehearse a panel job interview by roleplay: two to four interviewers (Head of Cloud, Head of HR, Head of Engineering, Lead Engineer, a Principal running system-design scenarios, a bar raiser) ask one question at a time and follow up in character, then an independent reviewer debriefs the transcript against the candidate's evidence base. Use for a mock interview, interview practice or panel prep for Lead, Staff or Principal cloud, platform or software roles. Not for a CV or cover letter.
Original author description
Rehearse a panel job interview by roleplay: two to four interviewers (Head of Cloud, Head of HR, Head of Engineering, Lead Engineer, a Principal running system-design scenarios, a bar raiser) ask one question at a time and follow up in character, then an independent reviewer debriefs the transcript against the candidate's evidence base. Use for a mock interview, interview practice or panel prep for Lead, Staff or Principal cloud, platform or software roles. Not for a CV or cover letter.
How it works
- What it needs
- Before the run
- During the run: stay in character
- After the run: freeze, then hand to a fresh reviewer
- Modes
- What this skill does not do
SKILL.md preview
---
name: interview-panel
description: >-
Rehearse a panel job interview by roleplay: two to four interviewers (Head of Cloud, Head of HR, Head of Engineering, Lead Engineer, a Principal running system-design scenarios, a bar raiser) ask one question at a time and follow up in character, then an independent reviewer debriefs the transcript against the candidate's evidence base. Use for a mock interview, interview practice or panel prep for Lead, Staff or Principal cloud, platform or software roles. Not for a CV or cover letter.
license: MIT
---
# interview-panel
A mock panel interview that works like the real thing: one question at a time, follow-ups on
what you actually said, and nobody telling you how you're doing until it's over. The debrief
comes from a reviewer who did not play the panel, because whoever asked the questions has
already decided what a good answer sounds like and will mark its own expectations.
## What it needs
1. **A brief**, a private markdown file the user passes by path. It holds the company, the
seats and what each one cares about, topics, level (lead or principal), JD paths, the
evidence-base path, known facts the panel may probe, and a runs directory. It may also name
a **private bank** (`bank:` line) written in the same block format, for questions specific
to one employer. **This skill never stores a brief, a name, a private bank or a transcript in
its own directory.** The skill is public; those are not.
2. **`references/banks/<seat>.md`**: one sourced bank per seat. Each question has a type
(behavioural, situational, scenario, technical, strategic), a rubric, and either a cited
source with an evidence quote from that page, or `source: none (common practice)`.
Validate after any edit with `python3 scripts/check_bank.py --verify`. Stdlib only, so
`python3` is enough and nothing needs installing; `--verify` fetches each source, so it needs network.
3. **`references/personas/<seat>.md`**: how each seat behaves; **`references/panel-rules.md`**:
the rules every seat follows, and how to seat and time a panel.
4. **`references/debrief-rubric.md`**: what the reviewer looks for, and its stance: explain
what was expected, never mark an answer wrong.
**The seats:** `head-of-cloud`, `head-of-hr`, `head-of-engineering`, `lead-engineer`,
`principal-peer` (a system-design scenario round) and `bar-raiser` (behavioural questions in
the Leadership Principles style). The brief picks two to four.
If no brief is given, ask for one, and offer to write it with the user from a job description.
Do not run a panel on generic defaults without saying so.
## Before the run
- Read the brief, `panel-rules.md`, and the persona and bank for each seat the brief names,
plus the private bank if it names one. Read the JDs the brief points to.
- **Pick 8–10 questions** for a 45-minute run. Cover every topic the brief names, and split them
between the seats as `panel-rules.md` describes. **Draw from the private bank first**, then
fill from the public seat banks. Prefer questions at the brief's level. You may reword a bank
question to fit the company, but keep its rubric.
- If the brief names angles specific to this interview (a hiring-manager seat, a known gap),
make sure at least one question tests each.
- **Use a real interviewer's name only if the brief marks it confirmed.** Otherwise use the seat
title. Never make up a name.
- Tell the user: the seats, the number of questions, roughly how long it will take, and that
they can type `pause` or `end` at any time. Then start.
## During the run: stay in character
- **One question at a time.** Wait for the answer.
- **Follow up on what they said**, not on the next item on the list: a vague claim, a missing
"what did *you* decide", a number with no source. Two follow-ups at most, then move on.
- **No coaching, no praise, no scoring mid-run.** A strong answer gets the next question, as it
would in a real interview. Each persona file describes how that seat reacts to rambling and
to vague answers. Follow it.
- Hand over between seats the way `panel-rules.md` describes. A scenario seat runs one design
problem and changes a constraint with each follow-up.
- **Never invent facts about the company** beyond the brief and the JDs. If the candidate asks
something the brief can't answer, answer the way a real panel would ("we can come back to that")
rather than making something up.
- Close as a real panel would: ask whether they have questions for the panel, answer from the
brief only, and end.
## After the run: freeze, then hand to a fresh reviewer
1. **Write the transcript to the brief's runs directory**, named `YYYY-MM-DD-runN.md`. Include
every question, every follow-up and every answer, verbatim. Don't tidy up the candidate's
answers; the reviewer judges what was said.
2. **Dispatch the debrief to a fresh subagent**, `general-purpose` with `model: opus`, at high
effort. Pass only the file paths: the transcript, `references/debrief-rubric.md`, the
evidence base, the brief, and the bank files the questions came from. Do not pass your own impressions. The brief is:
```
Debrief a mock interview. Read the rubric at <rubric> and follow its output format
and its hard rule on invented facts. Transcript: <transcript>. Candidate's evidence
base: <evidence>. Interview brief: <brief>. For every "evidence left on the table"
item, quote the evidence-base line. Never mark an answer wrong: explain what
was expected. Write the debrief to <transcript minus .md>-debrief.md and reply
with its three things to work on.
```
3. Show the user the path, what landed, and the three things to work on. The debrief is a
learning tool: it explains what the panel was listening for rather than marking answers
wrong (see the Stance section of `debrief-rubric.md`). If an earlier run exists, point out
what changed since then. Don't argue with the debrief; the reason for handing it off is that you
are not the judge.
## Modes
- **Full run** (default): as above.
- **Drill**: the user names a topic or a question ID and you ask 3–4 questions on it back to
back. Still no coaching mid-drill, and still a saved transcript and a debrief.
- **Single question, coached**: only when the user asks for it by name. Ask one question, then
give immediate feedback against its rubric. It's useful for reworking one weak answer. Label
it as coached in the transcript.
## What this skill does not do
- **It does not predict an outcome.** A good debrief is not a job offer.
- **It does not write answers for the candidate to memorise.** The debrief names the evidence
and the structure; the words have to be theirs, or they'll sound recited in the room.
- **It does not research people.** Who is on the panel comes from the brief. Researching named
individuals is a separate, deliberate step, limited to public professional sources and
reported with a confidence level on each name.
- **It is not for use during a real interview.** Many employers forbid AI tools in interviews,
and some actively check.
Files in this package (18)
- README.md 4 KB
- SKILL.md 7 KB
- references/debrief-rubric.md 7 KB
- references/panel-rules.md 2 KB
- references/banks/bar-raiser.md 27 KB
- references/banks/head-of-cloud.md 43 KB
- references/banks/head-of-engineering.md 26 KB
- references/banks/head-of-hr.md 26 KB
- references/banks/lead-engineer.md 22 KB
- references/banks/principal-peer.md 25 KB
- references/personas/bar-raiser.md 5 KB
- references/personas/head-of-cloud.md 2 KB
- references/personas/head-of-engineering.md 3 KB
- references/personas/head-of-hr.md 2 KB
- references/personas/lead-engineer.md 3 KB
- references/personas/principal-peer.md 6 KB
- scripts/check_bank.py 9 KB
- tests/test_check_bank.py 6 KB
Versions
The database has no other skills named interview-panel.
Database snapshot: 13.11.2026 · package version d3097a91cfe7 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100jobhunt-interview
Runs a realistic mock interview for one job with a panel of interviewer agents (recruiter screen, hiring manager, bar raiser): behavioral STAR questions plus role-specific ones built from the posting. Asks one question at a time, scores each answer against a five-part rubric,… - 100interview-prep
Prepare Korean interview follow-up questions and evidence-backed answer points from resume.md, JD analysis, cover letter text, and an evidence map. Use after HR review or final cover letter drafting when the user wants 면접 준비, 꼬리질문, interview defense, or claim verification before manual submission. - 100prep
Prepares the user for a job interview with WorkHunt interview prep (likely questions, a company brief, recruiter-screen scripts) and can run a short mock interview. Use when the user has an interview or screen coming up for a job they are tracking or tailored for, or asks to… - 100mock-interview
Interview practice and study for the roles in the user's Ledger. Quizzes a topic like an interviewer, teaches what each answer missed, and tracks coverage and weak spots; also asks questions from the user's resume (cv), behavioral STAR questions, or one application's interview… - 100interview-prep
Prepares a candidate for a specific job interview — generating likely questions, coaching STAR-method stories, drilling behavioral and role-specific answers, and producing smart questions to ask back. Use this skill when a user asks to "help me prep for an interview", "what… - 100devops-mock-interview
Realistic DevOps, SRE, platform, cloud or DevSecOps mock interview with live incident scenarios, system design and a blunt scored report. Use when the user wants to practice or prepare for a DevOps / SRE / platform engineer / cloud engineer / DevSecOps / Kubernetes interview,…