meeting-prep

What this skill does

# Meeting Prep

Research-powered meeting preparation with attendee intelligence and company context.

User request: $ARGUMENTS

**Before running any commands**, read `references/nimble-playbook.md` for Claude Code
constraints (no shell state, no `&`/`wait`, sub-agent permissions, communication style).

---

## Instructions

### Step 0: Preflight

Follow the transport selection + standard preflight from `references/nimble-playbook.md` — pick CLI or MCP at session start, then run the standard preflight calls (date calc, today, profile, memory index) in parallel.

From the results:
- CLI missing or API key unset → `references/profile-and-onboarding.md`, stop
- Tag all `nimble` CLI calls: `nimble --client-source skill-meeting-prep <subcommand>`. MCP path: not yet supported — see `references/nimble-playbook.md` for status.
- Profile exists → read `~/.nimble/memory/people/index.md` to identify existing
  person profiles. Load relevant `~/.nimble/memory/people/` files for attendees
  before — skip redundant searches, surface prior meeting notes. Follow
  `[[path/entity]]` cross-references in person files: if an attendee's file links
  to `[[competitors/widgetco]]`, load that competitor file for richer context (e.g.,
  recent intel from competitor-intel runs). Also check `~/.nimble/memory/companies/`
  for cached company research.
  **No same-day report check** — meeting-prep is per-meeting, not per-day. Users
  may prep for multiple meetings in one day. Instead, check entity freshness:
  if a person/company profile was updated within the last 24 hours, offer to reuse
  it: "I have a recent profile for **[Name]** from earlier today. Use it, or refresh?"
- No profile → that's fine. Meeting prep doesn't require onboarding. Proceed to Step 1.

### Step 1: Gather Meeting Context

Parse the meeting details from `$ARGUMENTS` or ask the user.

**Calendar shortcut:** If the user didn't specify attendees and a calendar connector
is available — either a calendar MCP tool (look for `list_events` in the tool list)
or the `gws` CLI (`gws calendar +agenda --today`) — offer to pull today's meetings
so they can pick one. If neither is available, skip this silently.

**If clear** (e.g., "prep me for my meeting with Alex Kim at WidgetCo tomorrow"):
- Extract: attendee name(s), company, meeting date/time (if given)
- Confirm briefly: "Preparing briefing for your meeting with **Alex Kim** at **WidgetCo**..."

**If partial** (e.g., "prep me for my meeting tomorrow"):
- Ask one clarifying question in plain text:
  > "Who are you meeting with? (names, titles, and company if you have them)"

**If just a person** (e.g., "research John Smith"):
- Proceed with the person. Try to infer their company from search results.

**Extract these fields:**

| Field | Required | Source |
|-------|----------|--------|
| Attendee name(s) | Yes | User input or calendar event |
| Company | Preferred | User input or inferred from search |
| Attendee title(s) | Optional | User input or discovered in Step 2 |
| Meeting type | Required | User input, inferred, or asked (discovery, demo, check-in, interview, partnership, internal) |
| Meeting date/time | Optional | User input |
| Additional context | Optional | User notes ("they're evaluating our product", "board member intro") |

**Meeting type detection** — if the user doesn't specify, infer from context clues:

| Signal | Inferred type |
|--------|---------------|
| "prospect", "demo", "sales call" | Sales / discovery |
| "interview", "candidate" | Interview |
| "board", "investor" | Board / investor |
| "partner", "integration" | Partnership |
| "check-in", "sync", "1:1" with colleague | Internal |
| No signal | Ask (see below) |

**If no signal** — don't guess "general external." The meeting type gates whether the
Value Positioning section is generated, so it's worth one question. Use AskUserQuestion:

> **What's the goal of this meeting?**
> - **Sales / discovery** — pitching, demo, exploring fit
> - **Partnership** — integration, co-selling, joint venture
> - **Board / investor** — board meeting, investor update, fundraising
> - **Interview** — evaluating a candidate
> - **General / other** — networking, catch-up, or not sure

Map the answer to the meeting type. If the user picks "General / other", treat as
general external (no value positioning section).

The meeting type shapes the briefing focus — specifically, it determines whether the
Value Positioning section (Step 4.5 + Step 6) is generated. Value positioning activates
for: **sales/discovery, partnership, board/investor**. It is skipped for: **interview,
internal, general external**.

### Step 2: WSA Discovery

Discover available WSAs for each attendee's company domain:

```bash
nimble agent list --search "{company-domain}" --limit 20
```

Run one search per unique company simultaneously. Filter for SERP/PDP WSAs,
prefer `managed_by: "nimble"`, validate with `nimble agent get --template-name {name}`.
Cache discovered names + params. Pass them to attendee agents in Step 3 for richer
data. If no WSAs found, continue with `nimble search` alone.

### Step 3: Per-Attendee Research (sub-agents)

Read `references/attendee-agent-prompt.md` for the full agent prompt template.
Follow the sub-agent spawning rules from `references/nimble-playbook.md`
(bypassPermissions, batch max 4, explicit Bash instruction, fallback on failure).

**Check memory first.** For each attendee, check `~/.nimble/memory/people/[name-slug].md`.
If a profile exists and is < 30 days old, load it as known context and pass it to the
agent so it focuses on what's new. If > 30 days old, run a full refresh.

Spawn `nimble-researcher` agents (`agents/nimble-researcher.md`) with
`mode: "bypassPermissions"`. One agent per attendee. Pass discovered WSA names
from Step 2 to each agent for enrichment.

**Important:** The Nimble API has a 10 req/sec rate limit per API key. With each agent
running 4-5 searches, limit concurrent agents to 2 per batch. For 3+ attendees, batch
in groups of 2.

**Call estimation & Scaled Execution:** Before launching agents, estimate total API
calls: ~5 searches per attendee + ~4 company searches + 3-5 extractions = ~(5 × N) + 9
calls. For 3+ attendees (15+ calls), tell agents to use `extract-batch` for page
extractions instead of individual calls. See the Scaled Execution pattern in
`references/nimble-playbook.md` for tier selection.

**Batch 1** (2 agents simultaneously):
- Attendee 1 research
- Attendee 2 research

**Batch 2** (if needed):
- Attendee 3 research
- Attendee 4 research

**Single attendee optimization:** If only one person, run the searches directly from
the main context instead of spawning an agent — saves overhead.

**Fallback:** If any agent fails or returns empty, run those searches directly from
the main context. Don't leave gaps in the briefing.

### Step 3.5: Gap Check

Before proceeding, verify every attendee has at least a title and company confirmed.

**For any attendee with < 3 meaningful results or "Role Unknown":**
1. Run a `--focus social` fallback search directly (this searches social platform
   people indices and is the most reliable way to find someone):
   `nimble search --query "[Name] [Company]" --focus social --max-results 5 --search-depth lite`
2. If `--focus social` is unavailable, fall back to:
   `nimble search --query "[Name]" --include-domain '["linkedin.com"]' --max-results 5 --search-depth lite`
3. Try name variations: "[First] [Last]", "[Full Name] [Company] [Title if known]"

Do NOT present a briefing with "Role Unknown" — exhaust social search first. If still
nothing after fallbacks, note it honestly: "Limited public presence — could not confirm
role. Consider asking for their LinkedIn URL."

Also collect **LinkedIn profile URLs** for each attendee during this step if not already
found. These are high-value for the final briefing output and Notion distribution.

### Step 4: Company Research

Research the attendees' company for meeting-relevant context. This is a