---

title: Setup and customization checklist description: Trigger setup, placeholders, policy decisions and the behaviour that must never be removed.

Setup and customization checklist

1. Trigger setup

The skill fires when an interview round closes for a candidate, or on demand ("debrief the panel for [candidate]"). Reference trigger:

Interview round [n] has closed for [candidate], [role].

Load and follow the <Panel debrief alignment> skill in full.

Context for this run:
- Candidate and role: [ATS link or ID]
- Interviewers this round: [names]
- Group debrief time, if one happened: [time]
- Who holds the candidate relationship: [name]

• {{ATS}} is connected and returns per-interviewer feedback with timestamps. The contamination check depends on the timestamps.
• Interviewers submit feedback in {{ATS}} before any group debrief. If your team debriefs verbally, change that first. The skill can flag contamination; it cannot remove it.
• Someone owns getting the candidate's read within 24 hours of the final interview.
• {{REVIEW_CHANNEL}} exists and {{HIRING_MANAGER}} actually reads it.

2. Placeholders

• {{ATS}}, {{NOTES_TOOL}}, {{REVIEW_CHANNEL}}, {{HIRING_MANAGER}}, {{COMPLIANCE_OWNER}}, {{DRIFT_WINDOW}} all filled, or the optional ones deleted.
• If you do not record interviews, delete {{NOTES_TOOL}} and its line in Step 1.
• Map "advance" to your own stage names in {{ATS}}.

3. Policy decisions

• Start in trial mode for the first three roles. ATS writes off, cards marked TRIAL.
• Drift window. 48 hours is the default. Shorter for SDR and high-volume roles, longer only if your process genuinely needs it.
• Who sees bias flags. Decide who {{COMPLIANCE_OWNER}} is before the first flag appears, not after.
• Work samples. Decide now whether you pay for them and the maximum time you will ask of a candidate. Never use a free strategy project.
• Brokering log. Decide where resolved concerns are logged so the library compounds across roles.

4. Behaviour that must never be removed

• BLIND rows are worked before DIVERGENT rows.
• Panel and candidate interest are rated separately, each with a quote.
• Feedback written after a group discussion is tagged and cannot make a row ALIGNED on its own.
• Missing inputs are named at the top of the card, never inferred.
• Interviewer words are never passed to the candidate verbatim.
• Protected-attribute comments are excluded from the decision and routed to {{COMPLIANCE_OWNER}}.
• No stage change, rejection or offer without a human Yes.
• Every ATS write is read back, and a failed write is reported on the card.