Prospect Research

Prospect Research — pre-meeting briefing notes

Produce one briefing note per target, optimized to be skimmed in ~30 seconds right before a meeting. Lead with memorable, human hooks; end with an explicit fit verdict. This is a decision-support note, not a neutral encyclopedia entry.

Scope (confirm per target)

Scope is set per name: ① person only ② company only ③ both. Use the scope given; if unstated, ask. "Company, recent-only" = skip the full history, cover recent developments.

The assessment lens (four axes)

Adapt the weights to the meeting's goal (sales / partnership / hiring / fundraising):

1. Fit / relevance — how well the target matches what this meeting is for (most important).
2. Capacity — budget, size, resources to actually act.
3. Decision power — decision-maker vs. gatekeeper vs. influencer.
4. Network value — who they can introduce, even if they don't act themselves.

Workflow (per target)

1. Web research (fan out with parallel subagents if available): bio, career, company, and relevance signals. Separate FACT from inference. Never fabricate — these are real people; "couldn't find it" beats a wrong fact. Treat subagent output as unverified leads, not facts: agents surface candidates + source URLs and can fail silently — re-verify before trusting.
2. Search internal records (do not skip): check your own CRM / notes / shared history for prior contact with the person AND the company. An existing relationship can flip the assessment entirely (already a customer, already in talks, a known intro path). Pull specifics from the actual records and fold them into the intro and verdict.
3. Verify each load-bearing claim yourself, then loop. Go bullet by bullet: for every claim that matters (title, tenure, size, a specific deal, identity) pull the primary source yourself — the profile, the company's own site, public registries. Confirm identity FIRST (common name → nail it via a second identifier) so you don't research the wrong person. Anything you can't confirm after a real attempt is marked (unverified) inline, or "couldn't find it" if blank. Repeat until every bullet is either primary-source-confirmed or explicitly marked.
4. Write the note in the format below.

Output format

One target = one note.

{Full Name}: {Company} ({Role})        ← profile link inline

<Intro: max 2 sentences. Role + one-line essence + why they matter to this meeting.
 If there is an existing relationship on record, lead with it.>

Person:
- age / location / background   (mark inference; drop the line if unknown and useless)
- Education:                    (its own line)
- Notable:                      (one memorable line)

Icebreaker:
- a genuinely non-obvious human hook; cross-link to others on the attendee list if there's a tie

Career:
- A → B → C …                   (keep company names; compress role descriptions)

Company ({Company}):            ← per scope
- what they do, founded, size (revenue / funding / headcount), notable points

Fit verdict:
- the four-axis read + any existing-relationship status. Bring real detail; don't defer with
  "confirm in the meeting". If they're the one selling to you, say so.

Style rules

• Lead with memorable hooks. The human angle is the point — the reader opens the profile himself, so don't just restate it.
• Intro ≤ 2 sentences. No filler framing.
• The verdict is a judgment, not a playbook. Cut multi-step "next actions" lists.
• Do the disambiguation; don't print the warning. Resolve same-name confusion silently, but keep real honesty: mark genuine uncertainty, and DO state a real risk finding.
• Education on its own line.
• Convert small foreign amounts to the reader's currency; large institutional figures can stay.

Verification mindset

Mark FACT vs. inference explicitly. Tie surprising/high-stakes claims to sources. Quotes must be real — check before quoting. Two traps that have shipped wrong notes before: (a) search-snippet mis-attribution — a snippet can splice another person's text onto your target, so confirm distinctive claims on the actual profile, not a snippet; (b) time-stamped marketing claims that go stale ("the only regulated X") — bind them to a date. It's fine to end with a short list of what's left unverified so the reader knows exactly what to confirm live.

Notes

Long-term context accumulates in the operator's own memory; this SKILL.md is the canonical procedure + output format. When feedback refines a rule, update this file so it persists.