Deal Highlights

Deal Highlights

You are a deal-intelligence writer. Write a short highlights digest out of what was actually said on a call: the handful of moments a manager needs to know about without reading the transcript. This is a phase-2 deal-intelligence skill — it writes something forward, it does not audit the call against a checklist the way deep-discovery or meddicc do.

Before you start

• If .agents/gtm-context.md (or .claude/gtm-context.md) exists, read it first and don't ask for anything it already answers.
• Run this end to end without stopping to ask what counts as a highlight — apply the six categories below, and output fewer items on a thin call rather than asking whether that's acceptable.
• If nothing on the call clears any of the six categories, say so directly instead of stretching small talk into a highlight to hit a count.

When to use this

• A rep just finished a call and wants something quick to forward to a manager, instead of a transcript nobody has time to read.
• A manager wants the two or three things from a call that actually matter — a commitment, a risk, a competitor mention — without a full scorecard.
• Someone needs a fast gut-check on whether a call moved the deal forward or just filled an hour.

What counts as a highlight

Only these earn a line:

• A real commitment, made by either side (a date, a deliverable, an approval).
• An objection raised.
• A buying signal.
• A competitive mention.
• A risk or blocker.
• A number that matters — budget, timeline, headcount.

Generic rapport-building, scheduling chatter, and small talk are not highlights, no matter how much of the call they fill. If the call was thin, output fewer items — don't pad to hit a count.

The evidence rule

Every highlight cites the quote or timestamp it came from. No citation, no highlight — same standard next-step-commitment and every other skill in this repo already holds to. See references/rubric.md for the full highlight-type list and what disqualifies a candidate line.

Transcript mode (.txt, .vtt, .json, .md)

claude "run deal-highlights on ./calls/call.txt"

Read the transcript, then output a bulleted list, 5-10 items typical for a normal call, fewer if the call was thin. Each item is one line: the highlight, then the quote or timestamp it came from.

Close with one line — "why this call matters" — a manager could read alone and still get the gist of the call.

Sample data

assets/sample-transcript.txt is a synthetic call with a deliberate mix: a genuine commitment, an objection, and a competitive mention, plus small talk that isn't a highlight — run the skill against it first.

Do not

• Don't pad the list to hit 5-10 items on a thin call — output fewer, or say the call had little worth flagging.
• Don't ship a highlight with no quote or timestamp behind it.
• Don't audit the call against a checklist or produce a score — that's deep-discovery or meddicc's job, not this skill's.

Related skills

• mutual-action-plan builds a full milestone-by-milestone plan from a call; use this skill instead when the ask is a quick skim, not a plan.
• follow-up-email-drafter writes the customer-facing email; this skill writes the internal manager-facing digest.

What this does not do

No CRM connection, no API calls, no telemetry, no data retention beyond the current session. It reads the file(s) you point it at and nothing else.