A Claude Skill that ingests an AE’s last ten calls from Gong and outputs a coaching note: three things working, two things to tighten, one specific exercise for next week. Built for managers who want consistent coaching but do not have ten hours a week to listen to calls.

What you’ll need

Gong with the call API enabled and call transcripts available
Claude.ai or Claude Code
Your team’s call rubric (discovery, demo, negotiation) as project knowledge
A weekly delivery channel agreed with the AE: DM, email, or in the next one-on-one

Setup

Install the Skill. ae-coaching.skill exposes pull_recent_calls, analyze_against_rubric, and render_coaching_note.
Auth Gong. API key with read access to the AE’s calls and transcripts. The Skill respects Gong’s permission model; if the AE’s calls are restricted, the Skill cannot see them.
Drop in the call rubric. Different rubrics for discovery, demo, and negotiation. The Skill auto-detects call type from the Gong stage tag or asks Claude to classify if untagged.
Set the rep ID and the date range. Default is the last ten calls or the last fourteen days, whichever is shorter.
Decide the delivery cadence. Weekly is realistic. Daily is too much. Monthly is too rare. Schedule for Friday afternoon so the AE has the weekend to absorb.

How it works

The Skill pulls call transcripts and metadata, classifies each by call type, and runs the relevant rubric against the transcript. Claude scores each rubric criterion, citing specific moments from the call. Then it aggregates across the ten calls into the three-two-one structure: three patterns working, two patterns to tighten, one specific exercise.

The exercise is the most important output. “Improve discovery” is useless. “On your next three discovery calls, ask the budget question before minute twenty” is actionable. The Skill is prompted hard to make the exercise concrete and measurable.

Watch-outs

Tone calibration. Coaching notes that read like a performance review get ignored. The Skill is prompted toward “trusted peer” voice. Test the tone with one of your strongest AEs before rolling out.
Transcript quality. Calls under five minutes, calls with more than five attendees, and bad-audio calls produce noisy analysis. Filter aggressively.
Rubric alignment. A rubric written for outbound discovery applied to inbound demo calls produces irrelevant feedback. Make sure the rubric matches the call type.
Manager-of-record dependency. This is a tool for the manager, not a replacement. The output is a draft. The manager edits, contextualizes, and delivers.

Stack

Gong — call source, transcript layer, type classification
Claude — rubric scoring and synthesis
Call rubric doc — the lens that makes feedback specific