call-prep

Call Prep

Produce a complete, scannable call prep brief by combining account research, contact research, and signal synthesis from Common Room.

Prep Process

Step 1: Identify the Account and Attendees

Parse what the user has provided:

• Company name — required; look up the account in Common Room
• Attendee names — optional; if provided, research each one

Calendar lookup (Lark): Pull the upcoming meeting from Lark Calendar to auto-fill attendees, time, and agenda. Use lark_calendar_agenda (project with jq — P3 — to the meeting whose title/attendees match the named company) to surface attendee names, the meeting time, and any agenda/description. If you only need to confirm a slot or the rep's availability around the call, lark_calendar_freebusy covers it. Use this to fill gaps the user didn't provide.

Resolve attendees (P1): For each external attendee name you find, resolve them with lark_contact_search(query="<name or email>") → open_id. You need the open_id to pull internal context (prior threads, shared meetings, past minutes) in the steps below.

If neither attendees nor a calendar match can be found, ask: "Who will be on the call from [Company]? I can research each attendee to make your prep more useful."

Step 2: Run Account Research

Use the account-research skill process to build a full account snapshot. For call prep, prioritize:

• Recent product signals (what are they doing in the product right now?)
• Open opportunities or renewal timeline
• Any risk signals (declining usage, support tickets, churned seats)
• Key recent events (funding, executive change, new hire)

When reviewing activity history, prioritize Gong and call recording activities — these provide direct context about previous conversations. Do not filter out call recordings by activity origin.

Prior Lark calls with this account (P6): Common Room captures third-party/community signals, but your team's own past calls with this customer live in Lark. Pull them as first-class prep context: lark_minutes_search(query="<company>", participant_ids="me") for the AI summary + action items of previous meetings, and lark_vc_search for the meeting envelope (who/when/room) when you only need metadata. Lead the brief with the most recent unresolved action items from these — they are the freshest, highest-signal talking points. Project reads with jq (P3).

Step 3: Run Contact Research for Each Attendee

For each external attendee, use the contact-research skill process. For call prep, focus on:

• Role and influence in the buying process
• Their personal activity and engagement history
• Any recent signals that suggest their current mood/priorities
• Spark persona classification if available

Step 4: Synthesize Talking Points and Objectives

Based on the combined account and contact research:

• Identify the call objective (e.g., discovery, demo, expansion conversation, renewal, QBR)
• Generate 3–5 tailored talking points grounded in specific signal data
• Anticipate 2–3 likely objections or topics the customer may raise
• Suggest a recommended outcome for the call

When the user's company context is available (see references/my-company-context.md), tailor talking points to the user's product and value proposition.

Step 5: Recency Check (Web Search)

After gathering all Common Room data, run a quick recency check to catch anything that happened since the last CR data sync. This is supplementary — CR data drives the prep; web search only adds recency.

Company news: Search "[company name]" news filtered to the last 14 days. Look for funding announcements, product launches, leadership changes, layoffs, partnerships, or press coverage.

Attendee presence: For each external attendee, search "[full name]" "[company name]" — look for recent articles, LinkedIn posts, conference talks, podcasts, or published opinions.

If a company news item is significant (e.g., just raised a round, announced a major hire), flag it in Signal Highlights. Otherwise, include findings briefly — don't let web search results overshadow CR signals.

Output Format

The output adapts to how much data Common Room returned. Only include sections where you have real data. Never fill a section with invented details.

When data is rich (multiple field groups returned, activity history, scores, signals):

## Call Prep: [Company] — [Date/Time if known]

**Meeting Context**
[Attendees, meeting type, and any known agenda]

---

### Company Snapshot
[4–6 bullets: key account status, signals, and recent activity]

---

### Attendee Profiles

**[Attendee Name] — [Title]**
[3–4 bullets: role, recent activity, Spark persona if available, personal hook]

[Repeat for each attendee]

---

### Signal Highlights
[Top 3 signals most relevant to this specific call]

---

### Talking Points
1. [Point tied to a specific signal]
2. [Point tied to a specific signal]
3. [Point tied to a specific signal]

### Likely Topics / Objections to Prepare For
- [Topic or objection + suggested response]
- [Topic or objection + suggested response]

### Recommended Call Outcome
[1–2 sentences: what success looks like for this meeting]

When data is sparse (few fields returned, no activity, null sparkSummary):

## Call Prep: [Company] — [Date/Time if known]

**Data available:** [List exactly what Common Room returned — e.g., "Name, title, email, two tags. No activity history, no scores, no Spark data."]

### What I Found
[Only the fields actually returned, presented as-is]

### Web Search Results
[Findings from web search on the company and attendees — or "No significant results"]

### Suggested Next Steps
- I can pull [specific field groups] from Common Room if available
- I can run deeper web searches on [specific topics]
- You may want to check Common Room directly for [what's missing]

Do not generate a full call prep brief from sparse data. A short honest output is always better than a long fabricated one.

Surface the brief as an interactive card (P4)

A call-prep brief is a digest the rep acts on right before a meeting — it belongs in chat as an interactive card, not a wall of text. When the rep wants it pushed to Lark (or it's part of a morning/weekly flow), post it via lark_im_card_send: a colored header (company + time), collapsible panel sections for Company Snapshot / Attendee Profiles / Talking Points, and an actions row ("Open calendar event", "Draft outreach"). Always print_json: true to validate the spec, then dry_run: true, then send. The plain-text format above remains the default for an in-chat answer.

Quality Standards

• Ground every talking point in a real signal — no generic filler
• Keep the brief tight — it should be readable in 5 minutes or less
• Flag unknowns explicitly — if attendee research is thin, say so
• Time-box the research — don't over-research at the expense of speed
• Never invent deal context — no fabricated proposals, competitor comparisons, pricing, trial terms, or objections not returned by a tool call

Lark-native handoff

Common Room owns the account/contact signal layer; Lark owns the collaboration layer this brief sits on:

• Calendar — lark_calendar_agenda / lark_calendar_freebusy for the meeting + slot (delegate depth to the lark-calendar skill).
• People (P1) — lark_contact_search to resolve every attendee to an open_id.
• Prior calls (P6) — lark_minutes_search (AI summary + action items) and lark_vc_search (envelope); delegate to the lark-minutes / lark-vc skills.
• Card (P4) — lark_im_card_send for the chat-native brief; card YAML grammar lives in the lark-im skill (references/spec-reference.md).
• See ../../connectors/LARK-PATTERNS.md, LARK-RECIPES.md, LARK-FUSION.md.

Reference Files

• references/call-types-guide.md — guidance for different call types (discovery, expansion, renewal, QBR) and how to tailor prep accordingly