A Claude Skill that ingests a rep’s calls, emails, and Salesforce activity for the last seven days and outputs a six-bullet signal report: what is heating up, what is cooling, where the rep is stuck, and one suggestion for next week. Designed for Friday self-review, not surveillance.

What you’ll need

Salesforce with Activity History enabled and API access for the rep’s user record
Gong access with the call transcript API turned on
Claude.ai or Claude Code
Your team’s qualification framework, dropped in as a markdown reference doc

Setup

Install the Skill. activity-summarizer.skill exposes three functions: pull_week_activity, pull_call_signals, and render_signal_report.
Auth Salesforce and Gong. Two tokens. Scope Salesforce to read-only on Activities and Opportunities. Scope Gong to read calls and transcripts the rep was on.
Drop in the qualification framework. MEDDPICC, BANT, your homegrown variant. The Skill uses it as the lens for “where is the deal stuck.”
Set the rep ID and the date range. The Skill defaults to the last seven days but you can pass a specific week.
Schedule for Friday at four. Output goes to the rep’s DM with a one-button “send to manager” follow-up.

How it works

The Skill pulls three streams in parallel. From Salesforce: every Task and Event the rep logged, plus stage changes on their open deals. From Gong: call summaries, talk-listen ratios, and any flagged objections. Claude then joins these against the open opportunity list and asks one question per deal: “What changed this week, and is it good?”

The output is intentionally short. Six bullets, max. Most weeks, three of them are accounts heating up, two are stuck deals, and one is a suggestion. Reps who get a forty-bullet activity dump will not read it.

Watch-outs

Privacy posture. This is rep-facing, not manager-facing. The output should land in the rep’s inbox first. If a manager wants a copy, the rep forwards it. Do not auto-cc.
Gong transcript quality. International calls and bad audio produce noisy transcripts that lead to bad signals. Filter calls under five minutes and flag anything below the confidence threshold.
Activity definition drift. What counts as a “meaningful touch” varies. Use your CRM’s task-type taxonomy and ignore the noise: ten “logged email” entries without a reply is not engagement.
Suggestion fatigue. The “one suggestion for next week” line drifts toward generic if the prompt is loose. Anchor it to an actual deal, an actual stage, an actual blocker.

Stack

Salesforce — activity and opportunity state
Gong — call signal layer
Claude — signal extraction and ranking against the qualification framework