Interview debrief summary with Claude

A Claude Skill that takes the scorecards and recorded transcripts from a candidate’s full interview loop and produces a structured debrief summary — synthesized evidence per structured interviewing rubric dimension, areas of agreement and disagreement across interviewers, and a draft hire/no-hire recommendation. Replaces the typical “everyone shares their gut feeling, loudest voice wins” debrief with evidence-grounded discussion.

What you’ll need

• Claude Code or Claude.ai with custom Skills enabled and Tier-A authorization for candidate-data handling per AI policy
• Scorecard data export from the ATS — Ashby, Greenhouse, or Lever
• Optional: interview transcripts from BrightHire or Metaview for richer evidence
• The role’s structured-interview rubric (the dimensions being evaluated)

Setup

1. Drop the Skill. Place interview-debrief-summary.skill into your Claude Code skills directory. The Skill exposes one callable function: summarize_debrief.
2. Configure the rubric. The Skill reads the role’s rubric from rubrics/{role_id}.yaml — the dimensions being evaluated, the score scale, anchor descriptions per level.
3. Test on a known candidate. Run on a candidate where the team has already debriefed. Compare the Skill’s summary to the team’s actual debrief notes. Tune the prompt if synthesis quality is off.

How it works

The Skill takes the candidate’s interview data and:

1. Loads scorecards across all interviewers. Pulls each interviewer’s scores per rubric dimension plus their written evidence/notes.
2. Loads transcripts (when available). From BrightHire or Metaview; finds the moments in the interview that support each scorecard claim.
3. Synthesizes per dimension. For each rubric dimension: aggregate score, range across interviewers (agreement vs disagreement), evidence supporting the score, evidence against, the moments in interviews where the signal appeared.
4. Drafts a recommendation. Hire / no-hire / hold based on the synthesized evidence. Explicit about confidence level and what would change the recommendation.

Output

A structured debrief document with five sections:

# Interview Debrief: [Candidate] — [Role]

## Aggregate scoring
- Overall: 3.7/5 (range 3-4 across 5 interviewers)
- Strong dimensions: technical depth (4.2), communication (4.0)
- Weak dimensions: systems design (3.0), execution under pressure (3.2)

## Per-dimension synthesis
[For each rubric dimension: aggregate score, range, evidence
for and against, supporting interview moments]

## Areas of disagreement
[Where interviewers diverged, the underlying disagreements,
the evidence each side cites]

## Recommendation
Hire / No-hire / Hold — with rationale and confidence level

## Suggested debrief discussion
- Top 3 questions the team should resolve in the debrief
- Specific rubric dimensions that need calibration discussion

Where it fits

Use this Skill before the debrief meeting, not after. The synthesis is the input to the debrief discussion — interviewers come into the meeting having seen the synthesized evidence and the points of disagreement, rather than spending the meeting comparing notes.

The downstream effect is that debriefs become 30-minute discussions of the actual disagreements rather than 90-minute reviews of who scored what.

Watch-outs

• Don’t replace the debrief discussion. The Skill produces evidence-grounded synthesis; the human team makes the hire/no-hire call. AI-recommended hires need human ratification.
• Rubric quality determines synthesis quality. A vague rubric produces vague synthesis. Codify rubric anchors before deploying.
• Sample-validate the recommendations. Periodically check whether the Skill’s recommendations align with the team’s eventual decisions. Systematic divergence reveals rubric or model issues.
• Privileged candidate data handling. Interview content is sensitive; verify the Skill operates within the firm’s AI policy for candidate data — Tier A enterprise AI only.
• Avoid surfacing demographic patterns. Some interview transcripts reveal protected-class information; the synthesis should not surface or use such information in the recommendation.