Interview Prep

---
name: metaview-interview-prep
description: >
  Build a complete pre-interview brief on a candidate from your Metaview interview history.
  Trigger when the user asks to prep for an upcoming interview or call with a named candidate —
  "prep me for my interview with Priya", "what do I need to know before my call with Marcus",
  "brief me on this candidate", "what's already been covered with Sofia", "help me not re-ask
  what the last interviewer asked". Pulls every prior conversation the candidate has had,
  their profile and scorecards, and turns them into a one-page brief: what's been covered,
  what's still unknown, contradictions to probe, and a tailored question plan for this round.
---

# Metaview Interview Prep

Walk into every interview already knowing what the last three interviewers learned. This skill turns a candidate's full Metaview history — transcripts, Notes, scorecards, profile — into a brief the interviewer can read in five minutes.

**Who it's for:** interviewers, hiring managers, and recruiters prepping the next round.
**Metaview products needed:** AI Notetaker (interview history). Scorecards/applications additionally need a connected ATS.

## Why this is powerful

Candidates repeat their story three times because interviewer #3 never saw rounds one and two — it reads as a disorganised company and wastes the round. The full history already exists in Metaview; this compresses it into five minutes of exactly what changes how you run the room: what's answered, what's contradicted, what's still unknown. Interviewers walk in further ahead than an hour of manual prep would get them, at the cost of one sentence: "prep me for my 2pm."

## Operating rules

- Tool names below may carry a connector prefix in your environment (e.g. `Metaview:search_conversations`). Match on the trailing name.
- Never show internal UUIDs (participant IDs, template IDs) to the user — refer to people and templates by name, and link conversations by URL.
- Everything in the brief must be grounded in the transcript, notes, scorecards, or profile you actually retrieved. If something wasn't discussed, say "not discussed yet" — that's a finding, not a gap in your work.
- This skill reads by default. The only write is the optional candidate pack in Step 7, which you offer rather than create unasked.

## Workflow

### Step 1 — Establish context (once per session)

Call `get_user_context`. Confirm the workspace name and note `data_access` — if the user can only see their own conversations, the brief may be missing rounds run by colleagues; say so in the output rather than presenting a partial history as complete.

### Step 2 — Resolve the candidate

Ask for the candidate's full name if not given (a first name alone usually collides).

1. `list_field_values` with `field_id="default:candidate"` and `search_term="<full name>"`. This searches external participants only — one entry per person, with name and email.
2. If several people match, disambiguate with the user by name + email domain before proceeding. The same human can occasionally appear as two records (e.g. personal + work email) — if so, gather conversations for both.
3. If nothing matches by name, try their email as the `search_term`, or ask the user for the Metaview conversation URL of any of their interviews.

Keep the returned `value` (participant ID) for the next steps — internally only.

### Step 3 — Pull their interview history

`search_conversations` with:

- `filters`: `[{"field_id": "default:candidate", "operation": "includes_one_of", "value": ["<participant-id>"]}]`
  (participant filters use `includes_one_of` / `excludes_all_of` — NOT `is_one_of`)
- Start with no extra fields to see `total_count` and the conversation list (dates, event titles, other participants), then choose depth:
  - **1–2 prior conversations:** re-query those IDs with `fields=["default:transcript"]` and read them fully. A one-hour transcript is roughly 10–15K tokens — budget accordingly.
  - **3–8 prior conversations:** re-query with `fields=["default:summary"]`, then pull the full transcript only for the single most decision-relevant round (usually the most recent, or the one covering this round's competencies).
  - **9+:** summaries only, and say in the brief which rounds you deep-read.

Note who interviewed them in each round and the round names (calendar event titles are often the best stage signal).

### Step 4 — Pull profile and ATS record

`fetch_candidates` with `[{"participant_id": "<id>"}]` (or their email). This returns:

- `profile`: experience, education, skills, location — useful for the snapshot header.
- `ats` (when connected and permitted): applications with current stage, **scorecards from prior rounds**, interviewer feedback, resume files.

Scorecards are the fastest way to see what prior interviewers concluded — read them against the transcripts: where a scorecard's claim isn't supported by anything in the transcript, flag it as worth validating rather than treating it as fact.

If `ats` is missing, that's normal (no ATS integration, or no permission) — build the brief from conversations + profile and note the limitation.

### Step 5 — Understand this round

Ask the user for the two inputs that sharpen everything: **the job description** (paste is fine — it anchors "gaps" to what the role actually needs, not generic competencies) and **what this round is meant to assess + its length**. If they can't say, infer from the application stage and what previous rounds covered, and state your assumption in the brief. If there's an interview guide for this stage, use its focus areas as the question plan's skeleton.

### Step 6 — Write the brief

Produce this structure, tightly:

```
# Interview brief — <Candidate> · <Round/purpose> · <Date if known>

**TL;DR** — 3 bullets: who they are, where the process stands, the one thing this round must resolve.

## Snapshot
Current role, tenure, location, headline experience. Salary expectations / notice period / work
authorization IF already discussed in a prior round (with which round it came from).

## What's been covered (don't re-ask)
Per prior round: interviewer, date, focus, and the 2-4 most decision-relevant things learned.
Link each round's Metaview URL.

## Signal so far
- Strengths raised by interviewers, quoted, with the round they came from
- Concerns / open questions with evidence
- Contradictions: places where rounds disagree, or scorecard says X but transcript shows Y

## Gaps this round should fill
The competencies/topics no one has probed yet, mapped to the role requirements the user provided.

## Suggested question plan
6-10 questions mapped to the gaps and concerns above — specific to THIS candidate (reference
their actual projects/claims), not generic. Mark 2-3 as must-ask.

## Candidate's side
Questions they've asked, what they care about, their stated timeline, other processes they
mentioned — so the interviewer can sell as well as assess.
```

Keep the whole brief readable in under five minutes. Cut anything that doesn't change how the interviewer runs the room.

### Step 7 — Offer follow-ups (don't do them unasked)

- Save the brief as a document (docx/PDF) or send it via a messaging connector if one is available.
- Generate a durable candidate pack inside Metaview: `generate_notes` on their most recent conversation with `template="candidate_pack"` — it lands in the Metaview app for the whole hiring team.

## Edge cases

- **First-round interview (no prior conversations):** the brief becomes profile + resume + JD driven. Say clearly there's no interview history; the "what's been covered" section becomes "what the application claims, to verify live."
- **Candidate found but conversations are 0 while ATS shows interviews:** the interviews may not have been recorded, or they're outside the user's data access. Report which.
- **Two candidates with the same name:** never merge them on a guess. Show both (name + email + last interview date) and ask.
- **The user asks minutes before the call:** lead with the TL;DR + question plan first, then fill in the rest.

## Example invocations

- "Prep me for my interview with Jai Singhal tomorrow — it's the final round."
- "What have we already asked Maria Duarte? I don't want to repeat the screen."
- "Brief me on the candidate I'm seeing at 2pm — Tom Okafor, product designer role."