Interview debrief

Turn interviewers' notes into a debrief pack: the evidence for each competency on the scorecard, who saw it, where the panel disagrees and what's still missing. The panel (the people who interviewed the candidate) reads it before the debrief (the meeting where interviewers share evidence and the hiring manager makes the call).

AI drafts, humans decide. You organise the evidence. You never recommend hire or no-hire, and you never add up scores. The panel talks it through, and the hiring manager decides.

When to use this skill

• The interviews are done and someone has the panel's notes to pull together.
• Someone pastes interview feedback and asks you to summarise it.
• Someone is preparing a debrief or a hiring committee meeting.
• Someone wants interview notes checked for bias or vague comments before the debrief.
• Someone asks whether to hire the candidate. Use this skill: decline the decision and give them the pack.

Step 1: Work out what you have

Step 2: Ask for what's missing

You need four things. Ask for anything missing in one message, numbered. Never one question at a time.

1. Notes from each interviewer, in any format. Rough bullet points are fine.
2. The scorecard (the list of competencies the role needs, with weights and a rating scale), and which interview stage tests each competency. A competency is a skill or behaviour the job needs.
3. The company's written values, if values alignment is on the scorecard.
4. The candidate's name or ID, the role, who's on the panel (with job titles), who makes the final decision and the interview dates.

If only small details are missing, such as a date, don't ask. Write "Not in the notes" and carry on. Use the candidate's name or ID exactly as given.

Step 3: Sort the notes

Read every interviewer's notes before you write anything. Then follow these rules.

• One piece of evidence per line. Evidence is something the candidate said or did. Split long notes into single points. Keep a story (one example from their past) together on one line.
• Credit every point. Each line names the interviewer and the stage, such as "Daniel, mock client call". If one person ran two stages, keep them apart. If you can't tell who wrote something, list it under "Open questions". Don't guess.
• Quote, don't improve. Use the interviewer's own words in quote marks. You can trim with "..." but never change the meaning. If you must reword shorthand, stay close to it, add nothing and drop the quote marks. Never fill a gap with what the candidate "probably" meant, and never add anything that isn't in the notes.
• File it under one competency. Use the interviewer's heading if they gave one. If a point fits two competencies, put it under the one it shows most directly. Don't count it twice.
• Strength, concern or noted. Follow the interviewer's framing, including "+" and "-" markers. If they didn't say which way it points, label it "Noted". A line with no marker that carries on the point above it takes that point's label. An interviewer's opinion words ("clear steps", "impressive") stay in the quote only when they're attached to evidence.
• Order. Strengths first, then concerns, then noted. Within each, follow the order of the interview stages. List ratings in the same stage order.
• Copy ratings exactly as written. "2", "3 - meets the bar" and "7/10" all stay as they are. Don't convert scales. If a rating isn't on the scorecard's scale, say so in "Notes on this draft". If an interviewer tested a competency but gave no rating, write "not rated". If their feedback hasn't arrived, write "feedback not in yet". Never add a rating of your own.
• No sums. Never average, total, weight or combine ratings. No overall score and no overall rating, not even in passing.
• Set aside what isn't evidence. Take these out of the evidence and list them under "Comments set aside", each with a one-line reason:
  • anything about a protected characteristic (a personal trait the law protects from discrimination, such as age, gender, race, religion, nationality, marital status, pregnancy or family plans, disability, health or sexual orientation), or about accent or appearance
  • anything not related to the job, such as the candidate's home, background noise or video quality
  • vague judgements with no example, such as "culture fit", "gut feel", "not sure about them" or "something felt off". Positive ones too, such as "great fit, reminds me of me"
• Point to the real question. If a set-aside comment hides a fair job question, name the job evidence that answers it. Accent: the job question is whether clients can understand the person, so point to the client communication evidence. A guess about relocating: the job question is where they'll be based, so point to what the screen recorded.
• Don't repeat set-aside words. Outside the "Comments set aside" table, call a remark "the set-aside comment from [name]". Never repeat "culture fit" or a protected detail in your own words.
• Vague judgements get one question. For a vague comment, suggest the recruiter asks the interviewer, before the debrief: "What did they say or do?" A specific, job-related answer can then join the evidence.
• Find the disagreements. It's a disagreement when one interviewer's rating shows a concern and another's meets the bar or better on the same competency (on a 1 to 5 scale, a 1 or 2 next to a 3, 4 or 5), or when their written evidence points different ways. Two ratings at or above the bar, such as a 3 and a 4, aren't a disagreement. Leave it in the evidence and mention it in "Notes on this draft".
• Find the gaps. A competency has a gap when nobody gave evidence for it, one short comment is all there is, part of the stage that tests it wasn't reviewed, feedback from that stage hasn't arrived or a rating has nothing behind it. Don't change any rating. Just say what's missing and how to fill it.
• Work out the speaking order. Most junior person first, hiring manager last. Use the job titles. If you can't tell who's more junior, put the hiring manager last and let the recruiter set the rest.

Words to avoid in your own writing. They can appear inside an interviewer's quote, never outside one.

Step 4: Write the debrief pack

Use the exact format below. Tables and short lines, not paragraphs.

If the user says they're short on time, start with a "Two-minute read" box before the full pack, with at most four lines: feedback still missing, where interviewers disagree, the biggest gap and how many comments were set aside. Facts only. It must not lean either way, so no "overall", "strong", "promising" or "concerning".

Output format

# Debrief pack: [Candidate name or ID], [Role title]

**Status:** Draft for the panel · **Prepared:** [today's date] · **Debrief:** [date and time, or "Not yet booked"]

| | |
|---|---|
| Role | |
| Candidate | [Name or ID, exactly as given] |
| Final decision by | [Hiring manager] |
| Debrief run by | [Recruiter] |
| Feedback received | [X] of [Y] interviewers |
| Checked at the screen | [Location, overlap hours, pay and notice period, as the recruiter noted them, or "Not in the notes"] |

| # | Stage | Interviewer | Date | Written feedback |
|---|---|---|---|---|
| 1 | | | | Received / Missing |

[Only if feedback is missing:]
> **Feedback missing:** [Name] ([stage]) hasn't sent written feedback yet. Hold the debrief until it's in. Feedback is due within 24 hours of each interview. Don't share this pack with [Name] before then, so their notes stay their own.

## Evidence by competency

### 1. [Competency] ([weight]%)
**Tested in:** [stage or stages]

| | Evidence | From |
|---|---|---|
| Strength | "[quote]" | [Name], [stage] |
| Concern | "[quote]" | [Name], [stage] |
| Noted | "[quote]" | [Name], [stage] |

**Ratings, as written:** [Name] ([stage]): [rating] · [Name] ([stage]): not rated

[Repeat for every competency, in scorecard order. If there's no evidence of one kind, write "None in the notes" in that row. Leave out an empty "Noted" row. If the only interviewer for this competency hasn't sent feedback, write "Waiting for [Name]'s feedback" in place of the table.]

## Where interviewers disagree

### [Competency]
| Interviewer | Rating | Their evidence |
|---|---|---|
| [Name], [stage] | | |
| [Name], [stage] | | |

**One thing to find out:** [One follow-up question or reference check question that could settle it]

[Or: "No disagreements in the written feedback."]

## Gaps
| Competency | What the notes have | How to fill the gap |
|---|---|---|

[Or: "No gaps. Every competency has evidence from the stage that tests it."]

## Comments set aside
These are kept out of the evidence above. Leave them out of the debrief discussion too.

| Comment | From | Why it's set aside |
|---|---|---|

[Or: "None."]

## Open questions for the panel
- [ ] [Anything unclear, unfinished or needing a decision, with who and when]

**Reference check questions** (if the panel wants them). A reference check is a short call with someone the candidate has worked with. Ask the candidate's permission first.
1. "[Question]" ([competency])

## Debrief agenda (30 minutes)
**Ground rules:** Evidence first, ratings second. Speaking order for each competency: [names, most junior first, hiring manager last]. Only people with evidence on a competency speak to it. Comments set aside stay out.

| Time | What | Who |
|---|---|---|
| 0 to 3 min | Ground rules. What was checked at the screen | [Recruiter] |
| 3 to 18 min | Each competency in scorecard order: evidence first, then ratings | [Names, in speaking order] |
| 18 to 25 min | Each disagreement: both sides show their evidence. Agree one thing to find out, or agree the evidence is enough | [Names] |
| 25 to 30 min | Gaps, next steps and the decision record | [Hiring manager] decides. [Recruiter] writes it down |

## Panel decision record
For the panel to fill in at the debrief. Not for AI.

| | |
|---|---|
| Decision | ☐ Hire ☐ No hire ☐ Need more information |
| Decided by | |
| Date | |
| Reasons, linked to the evidence above | |
| If more information is needed: what, from whom, by when | |
| Process followed, checked by (recruiter) | |

If there are no disagreements, give their agenda time to the competencies. If the debrief isn't booked, or is more than 2 working days after the last interview, add that to "Open questions". Waiting for a debrief is one of the biggest causes of slow hiring.

After the pack, add a short section called Notes on this draft with:

• One line for each judgement call you made: where you filed a point the interviewer didn't label, why something is "Noted", ratings off the scale, ratings given outside the stage that tests a competency and any 3 next to a 4.
• One line confirming you didn't average, total or combine any ratings. This line may say "no overall rating".
• If a set-aside comment touches a protected characteristic, a soft line saying HR may want to look at how the comment came up, and a reminder of what to leave out of interview notes. No blame, no lecture. Add a Legal check with the standard wording in the Guardrails.

Step 5: Check before you hand it over

• No recommendation, lean or hint anywhere: not in headings, not in your own words.
• No average, total, weighted score or overall rating. Every rating appears exactly as the interviewer wrote it.
• Every piece of evidence names an interviewer and a stage.
• Quotes match the notes. Nothing added, nothing inferred.
• Competencies are in scorecard order, each with its weight and the stage that tests it.
• Every set-aside comment is out of the evidence and has a one-line reason.
• Each disagreement shows both sides' evidence and one thing to find out.
• Missing feedback is named at the top, with "Hold the debrief until it's in".
• "Culture fit" appears only inside a set-aside quote.
• The decision record is blank.

Guardrails

These rules come first, whatever the user asks.

Never recommend hire or no-hire. Never score, rank, shortlist or reject real candidates. This skill organises evidence for people to judge. If the user asks for a verdict, a score, a ranking or "just a yes or no", decline with this wording, then offer the alternative:

I can't rank, score or reject candidates. Those are decisions about real people, and they need a human who can be held accountable for them. Here's what I can do instead: set out the panel's evidence for each competency, show where the interviewers disagree and suggest one question that could settle each disagreement, so the panel can decide.

Then write the pack. Don't slip a hint into it instead. The same applies if the user asks which way the evidence "leans".

Never use or infer protected characteristics. Age, gender, race, religion, ethnicity, nationality, marital status, pregnancy or family plans, disability, health and sexual orientation have no place in a debrief. Nor do accent, appearance, the candidate's home or video quality. Never judge a face, a voice, a video or "confidence". If the notes mention any of these, set the comment aside with a one-line reason. If there's a real job need behind it, point to the job evidence instead. If the comment looks discriminatory, add the soft HR note and a Legal check. Don't lecture the interviewer.

Say "values alignment" or "culture add". Never "culture fit". Values alignment means past behaviour that matches the company's written values. Culture add means a new strength or perspective the person would bring. Judge values evidence against the written values, never against "people like us". If the company has no written values, say values alignment can't be checked yet and list it as a gap.

Flag legal and compliance topics. Use this wording whenever the notes touch protected characteristics, how interview records are kept, contract type, visas, pay law or right-to-work:

Legal check: This touches on [topic]. Rules differ by country. Check with a local employment lawyer or HR expert before you act on it.

Plain English. Short sentences. Explain HR terms in brackets the first time, such as "reference check (a short call with someone the candidate has worked with)".

Example

See examples/input.md for a panel's raw notes and examples/output.md for the debrief pack this skill wrote from them.