Playbook

Playbook: Build a Discovery Call Summary Skill

It is Tuesday afternoon. Four AEs on the sales team just got off discovery calls. All four are writing summaries. One pastes the transcript into their person...

Agent Skillsagent-skills / playbook-discovery-call-summary

It is Tuesday afternoon. Four AEs on the sales team just got off discovery calls. All four are writing summaries. One pastes the transcript into their personal ChatGPT and asks for "the key takeaways." One uses a prompt from a doc the sales manager wrote eighteen months ago that still asks the model to identify "pain points and budget." One has their own template they refined on Sunday over coffee. One does not write a summary at all because they "remember what was said." By Friday, the CRM has four notes in four formats, none of them flag the same risks, and the AE running the next call with the same prospect cannot tell what was already discussed.

This is the sales version of the scattered-AI problem. The work is happening. It just isn't owned. This playbook walks you through replacing those four prompts with one reviewed Agent Skill that produces a consistent discovery call summary — qualification fields filled in, risks flagged, next steps explicit — attached to your sales agent so every AE produces the same shape of summary regardless of which prompt they used to write before.

What you will build

By the end of this playbook, you will have:

  • A workspace Agent Skill named Discovery Call Summary with a SKILL.md procedure, the qualification framework your team uses (MEDDIC, BANT, Force Management, or your own variant), a CRM-ready output template, examples of good and bad summaries, and the rules for what gets flagged for manager review.
  • An approved version of that skill, attached to your Sales Agent with a readable mount path.
  • A repeatable workflow: an AE drops a call transcript or notes into the sales agent, gets back a structured summary that maps to the CRM fields, and the manager pipeline review meeting works because every account is described the same way.
  • A versioning path so the qualification fields can evolve as the team's methodology evolves, without breaking last quarter's notes.

What you need before you start

  • An Assist workspace where you can create workspace-scoped Agent Skills.
  • A workspace admin who can review and approve versions.
  • A Sales Agent in your workspace — the operational agent your sales team uses, ideally with tools that can write to the CRM (Salesforce, HubSpot, Attio, whatever you actually use). If you do not have one, create a subagent first.
  • A sales manager and one or two senior AEs willing to define the qualification framework and validate the skill before rollout.
  • Three to five real (sanitized) discovery call transcripts or detailed notes for validation. Mix the scenarios: a strong qualified deal, a weak fit, a competitive situation, an inbound that turned out to be a researcher.

Step 1: Collect the scattered prompts and the four formats

Ask each AE on the team to share what they actually do today. Not what they think they should do — what they do.

"Send me the prompt you use to summarize a discovery call. Copy it from your personal account or doc. Also send me the last discovery summary you wrote — the actual text that went into the CRM, not a cleaned-up version."

Collect four to six samples. Patterns will surface immediately. Most prompts ask for some flavor of takeaways. Some have qualification questions embedded; some don't. Some name the prospect company in the prompt; some include direct quotes from the buyer with attribution. The CRM notes will be even more telling — different headings, different fields, different levels of specificity.

Save the source prompts and the source CRM notes in a draft doc. They become the input for authoring. You will point at specific failures: "this prompt told the model to identify pain even when the call was discovery without a pain conversation" or "this CRM note has a direct quote with the buyer's name in plain text."

Step 2: Lock down the qualification framework first

Discovery call summaries are only useful if the team agrees on what gets summarized. Before you author any skill files, get the sales manager and one or two senior AEs in a room and answer: what are the exact fields every discovery summary must populate?

A reasonable starter set for B2B sales:

  • Account name and stage in the pipeline
  • Primary contact (name, role, decision authority)
  • Other stakeholders mentioned on the call
  • Business outcome the buyer is trying to achieve
  • Current solution or workaround (what they use today)
  • Compelling event or timeline driving urgency
  • Budget signal (named, ranged, unknown)
  • Competitive landscape (who else is in the deal)
  • Identified risks
  • Agreed next steps with owners and dates
  • Manager-review flag and reason, if applicable

Write the field list down. Get the manager to sign off on it before you write a single skill file. The fields are what the agent fills in — if the team disagrees on the fields, the agent cannot make them agree.

Step 3: Create the draft skill

Open Assist and go to Workspace > Agent Skills. Click New Skill and set:

  • Name: Discovery Call Summary
  • Slug: discovery-call-summary
  • Scope: Workspace
  • Description: Produces a structured discovery call summary that maps to the team's CRM fields, applies the qualification framework, and flags accounts requiring manager review.

Or scaffold it from chat:

"Create a workspace Agent Skill called Discovery Call Summary. Slug: discovery-call-summary. It produces a CRM-ready discovery call summary using our qualification framework and flags accounts that need manager review. Start a draft SKILL.md and folders for templates, policies, and examples."

The skill is a draft. Nothing is mounted yet.

Step 4: Write the main procedure

Open /SKILL.md. The first line tells Assist when to pick this skill up:

Use this skill when an AE asks to summarize a discovery call, a first prospect conversation, or an early-stage sales call for the CRM.

Then walk through the procedure. A useful SKILL.md for this skill includes:

  1. Confirm the account name and the call source: transcript, notes, or recording transcript. Refuse to summarize a "discovery call" with no input — hallucinated summaries in the CRM corrupt pipeline reviews.
  2. Read the full input. Identify the primary contact, their role, and their decision authority based on what they actually said. Do not infer authority from titles alone.
  3. Populate each field in the qualification framework using the template in /templates/discovery-summary.md. If a field cannot be populated from the call, mark it as "Not discussed" — never fabricate.
  4. Apply the customer-data and quote policy in /policies/customer-data-rules.md. Direct quotes from the buyer require attribution; sensitive figures (revenue, headcount) are stated as ranges if mentioned, never as precise numbers unless the buyer explicitly confirmed them.
  5. Identify risks from the patterns in /policies/risk-flags.md: no compelling event, no economic buyer involved, single-threaded, competitor is incumbent, custom commitments offered by the AE.
  6. List agreed next steps with named owners and dates. Steps without an owner or date are not next steps — they are wishes.
  7. Set a manager-review flag if any of these conditions are true: deal size above the manager's review threshold, a custom commitment was made on the call, the buyer mentioned legal or procurement involvement, or two or more risk flags fired.
  8. Produce the output in the strict format the template defines.

The procedure is what stops the agent from producing a "professional-sounding" summary that misses the qualification fields the manager actually reviews.

Step 5: Add the supporting files

  • /templates/discovery-summary.md — the strict output format, with every qualification field as a required heading. Empty fields are explicitly marked "Not discussed" — never silently omitted.
  • /policies/customer-data-rules.md — direct-quote rules, sensitive-figures rules, contact PII rules. Sales summaries that leak buyer-side detail without consent are a real customer-trust risk.
  • /policies/risk-flags.md — the patterns that count as risk for your sales motion. Be specific: "no compelling event" means the buyer did not name a date or event driving urgency; "single-threaded" means only one buyer-side stakeholder is engaged; "custom commitments" means anything outside the standard deal terms.
  • /examples/good-summary.md — one fully worked example summary, with all fields populated correctly, risk flags applied, and a manager-review flag with reasoning. Show what "Not discussed" looks like for a field that legitimately didn't come up.
  • /examples/bad-summary.md — examples of the failure modes: fabricated budget, inferred authority that wasn't said, missing next-step owners, wishful summaries that read like the AE's hopes rather than the buyer's words.

Drive the work from chat:

"Draft /policies/risk-flags.md for the Discovery Call Summary skill. Include the eight risk patterns: no compelling event, no economic buyer engaged, single-threaded contact, incumbent competitor with strong satisfaction, custom commitments by the AE, vague timeline, missing budget conversation, and procurement-driven evaluation. For each, define the evidence the agent should cite."

Review every file before submitting. The output is the file. If the file is vague, the output is vague.

Step 6: Submit for review and approve

From the skill detail page, click Submit for review. The sales manager and a workspace admin walk through the files:

  • Does the qualification framework match what gets reviewed in pipeline meetings?
  • Do the risk flags reflect how the manager actually thinks about deals?
  • Does the customer-data policy match how the company wants buyer-side detail handled?
  • Does the good-output example match what a pipeline review meeting would want to see?

When approved, the version is available to mount. If rejected, the feedback comes back as comments on the version. Author a new draft addressing the points — do not negotiate the rejection in chat. The version is the contract.

Step 7: Attach the skill to the Sales Agent

Open the Sales Agent and attach the approved version with a readable mount path:

/skills/discovery-call-summary

Now every AE who asks the Sales Agent to summarize a discovery call gets the reviewed workflow. The four personal prompts and the eighteen-month-old manager doc become irrelevant.

Step 8: Validate on real calls

Pick three to five real (sanitized) discovery call transcripts. Run them through the Sales Agent:

"Summarize this discovery call with Acme Corp. The AE was Priya. Here's the transcript."

"Produce a discovery summary for this call. The buyer mentioned procurement involvement halfway through."

"Summarize this discovery call. It's a competitive situation — the buyer is currently using a competitor and the contract renews next quarter."

The sales manager and a senior AE read the output and check:

  • Are all qualification fields populated correctly, with "Not discussed" used honestly?
  • Did the agent flag the right risks? In the competitive example, did "incumbent competitor" fire?
  • Did the manager-review flag fire when it should have — did the procurement signal in the second example trigger it?
  • Does the format match what the pipeline review meeting actually uses?

Iterate by updating the skill files. If the agent flagged "single-threaded" for a call where two stakeholders were genuinely engaged, the risk-flag rule needs to be more precise about what counts as engagement. If the agent fabricated a budget figure, the customer-data policy needs to forbid budget inference more explicitly.

Step 9: Roll out to the sales team

Once validation passes, announce the change. Tie it to the pipeline review:

"Starting next week, all discovery summaries are produced by the Sales Agent using the reviewed Discovery Call Summary workflow. The fields match what we review in pipeline meetings. The risk flags match how the manager thinks about deals. Stop using your personal accounts for this — the agent now produces a better summary in the format the team uses."

Stop maintaining the old manager doc. Link AEs from any old reference to the agent. The personal prompts atrophy because the agent's output is genuinely better — and consistent with what gets reviewed.

Step 10: Improve the skill from real calls

After two weeks of real use, the patterns surface. Maybe the agent is under-flagging deals because the risk thresholds are too high. Maybe the team realized "compelling event" needs sub-categories (regulatory deadline vs internal initiative vs competitive renewal). Maybe pipeline review identified a new field everyone wants: "internal champion strength" on a 1-3 scale.

Open a new draft of the skill, fix the files, submit for review. When approved, deliberately upgrade the Sales Agent's attachment using the workflow in Governing Agent Skill versions. Existing summaries don't get rewritten — old summaries reflect the framework as of when they were written, which is exactly right for historical pipeline data.

What you built

You have a reviewed Agent Skill that produces discovery call summaries the way the sales team actually wants discovery call summaries to look. The skill has:

  • A qualification framework everyone agreed on, populated by the agent the same way every time.
  • A customer-data policy that applies to every summary, not just the ones written by the careful AE.
  • Risk flags that match how the manager evaluates deals in pipeline review.
  • A manager-review trigger that fires on conditions the manager actually wants to see flagged.
  • An output format that maps directly to the CRM fields and the pipeline review meeting.
  • A clear path to improve the skill as the sales motion evolves, with version control and announced upgrades.

More importantly, you replaced four personal AE prompts and an eighteen-month-old manager doc with one reviewed workflow the sales team owns. The pipeline review meeting becomes more useful because every account is described in the same shape. The next new AE joining the team does not invent their own discovery summary prompt — they use the sales agent. The next time procurement gets mentioned on a discovery call, the manager-review flag fires whether the AE noticed it or not.

This is what bringing scattered AI use under one platform looks like in the sales function: one approved skill, one operational agent, one consistent summary, with the qualification fields actually populated.

Where to go from here

  • Build a deal review prep skill. The natural next workflow: take the discovery summary plus subsequent call summaries and produce a deal review packet for the manager. Same playbook, different qualification framework (closer to MEDDIC at this stage).
  • Wire the agent to write directly to the CRM. If the Sales Agent has CRM tools, the discovery summary can land directly in the opportunity record, populating the fields. The output template you wrote becomes the field mapping.
  • Add a sandcastle pipeline view. Build a sandcastle that shows the team's pipeline grouped by risk flag, manager-review flag, and last-discovery-summary date. The summaries this skill produces are the input.
  • Quarterly framework drift review. Sales methodologies evolve. Have the sales manager re-run validation prompts each quarter against the framework and update the skill when the framework changes — not in chat, in the reviewed file.