Discovery Call Master — System Guide

The guide

A Claude Code skill that turns a discovery call transcript into structured notes, an ICP score for the prospect, a CRM row update, new insights in your strategy doc, web-researched competitor entries (the skill fills both your competitor narratives and your comparison table), and product-backlog updates from every feature request, unmet need, or workaround the prospect mentioned. All in one pass, all stored as plain Markdown in your repo.

The system at a glance

TRANSCRIPT
   │
   ▼
SKILL  ←──reads──  CLAUDE.md  (your business context + ICP rules)
   │
   ▼
ONE PROPOSAL MESSAGE  (you validate or correct, per category)
   │
   ▼
PARALLEL WRITES INTO YOUR REPO:
   ├─ Structured call file  (frontmatter + key learning + summary + raw transcript)
   ├─ CRM row updated  + ICP score with one-line justifications per criterion
   ├─ Insights routed  to the right section of your strategy doc
   ├─ Competitor entries  researched on the web → narrative file + comparison table
   └─ Product backlog  filled with feature requests, unmet needs, workarounds

What matters in this system:

• Everything is plain Markdown in your repo. Version-controlled, greppable, yours. No SaaS, free.
• Validation before writing. The skill proposes everything in one message. You correct or confirm per category. It writes nothing without your green light.
• Web research is real. For each new competitor mentioned, Claude searches founders, funding rounds, pricing, customer segments, and writes both a narrative entry and a row in your comparison table. Not just a name on a list.
• The skill is only as good as your CLAUDE.md. Company context and ICP rules are what let it score, route insights, and judge competitor relevance. This file is the brain.

What you need

• Claude Code installed. https://docs.claude.com/claude-code
• A repo for your business docs. Markdown + CSV.
• A CLAUDE.md at the repo root with two parts:
  • Company description — what you build, target customer, market, known competitors, tone of voice.
  • ICP definition — criteria + scoring scheme (a scale and a rule per criterion).
• A transcript source. I use Granola free tier (calls deleted after 30 days, doesn't matter — you process them the same day). Fathom, Otter, Zoom transcripts, or typed notes work too.

Files the skill reads and writes

You can rename or move any of these by editing the paths inside SKILL.md before installing.

The 5 actions, in detail

1. Structured call file

The skill creates a Markdown file at discovery-calls/YYYY-MM-DD-name-company.md with a YAML frontmatter (date, participant name, role, company, LinkedIn, call type, interest level, topics) followed by five sections in this order: the highlight learning of the call, the concrete next steps, a structured summary covering the prospect's context and current tools and real needs, a clean version of your raw notes, and the raw transcript pasted as-is. This file becomes the source of truth you grep weeks later when you prep a follow-up or look back at what was said.

2. CRM row updated + ICP score

The skill reads the current header of your CRM and proposes either a new row (new contact) or an update (existing contact), mapping fields from the call file to your columns. The ICP score is computed by reading the ICP rules in your CLAUDE.md and applying them to what the prospect said in the call: each criterion gets a one-line justification, and the final score uses the scale you defined (1-5, A-D, 0-100, whatever). The justifications appear in the validation message so you can correct them on the spot before they land in the CRM. The point is to make every prospect comparable later when you sort your pipe.

3. Insights routed to your strategy doc

For each insight worth keeping (something the prospect said that updates your view of the market, the buyer, or the product), the skill reformulates it cleanly without changing the substance, attaches a source (First name — Company), reads the existing sections of insights.md, and proposes the most relevant one for placement. If no existing section fits, it proposes a new short section name. Before writing, it checks for near-duplicates and proposes enriching an existing insight with the new source rather than creating a duplicate. Without this routing logic, insights pile up in one big bucket and the doc becomes unusable; section-aware placement keeps it queryable as it grows.

4. Competitor entries (web research → narrative file + comparison table)

For each competitor or tool the prospect mentioned (interpreted broadly: anything in their workflow that touches your market, not only direct competitors), the skill checks if it is already documented. If not, it runs a web search to gather founders and their backgrounds, founding year, location, public funding rounds and main investors, pricing model, customer segments, and traction signals. It then writes a narrative entry in competitors.md with three sections (What they do / Their story / Numbers) and adds a matching row in competitors-comparison.md using the columns you already defined. If you keep a TSV mirror of the comparison table, that gets synced too. Most discovery calls surface 2-3 tools you have never heard of or only know by name; this turns each mention into a researched entry, not just a name on a list.

5. Product backlog

For each signal of unmet need the prospect surfaces, the skill writes an entry in product-backlog.md with a one-line title, a short description, the source (First name — Company), the context in which it came up, and a tag: explicit-request if they asked for it directly, implicit-need if they described a hard part of their workflow without framing it as a request, or workaround if they described a manual or expensive process they run today that your product could automate. The tag matters because the three categories are not equal: explicit requests are easy to dismiss, implicit needs sit between two products, and workarounds are usually where the real product opportunity lives.

What you need to adjust in SKILL.md

The skill ships with defaults that match my own setup. Some of those will not match yours. Scan these places before you run it on a real call.

File paths

The skill expects discovery-calls/, crm.csv, insights.md, competitors.md, competitors-comparison.md, product-backlog.md. If your business docs live in a different structure (for example docs/sales/crm.tsv or docs/strategy/insights.md), open SKILL.md and replace the paths before installing. The skill reads paths from one place; one search-and-replace pass is enough.

CRM column names and mapping

SKILL.md carries a default list of CRM columns and a mapping from each field of the call file to a CRM column. If your CRM uses different column names, edit the Mapping from the notes file block in Phase 2 to match yours. The skill reads your header at runtime, but the mapping logic still needs to know which call field belongs in which CRM column.

Summary section structure in Phase 1

The default Summary section is written for sales discovery (their context, current tools and limits, real needs, expectations for a new tool, go-to-market and adoption context). If you run customer research or product interviews instead, replace those subsections with what fits your framework — for example: their goal, current workflow, friction points, what they tried, what would make them switch.

Comparison table columns

Your comparison table reflects your product's differentiation axes, which are unique to you. The skill reads the header of competitors-comparison.md at runtime, so you do not edit the skill itself — but you do need to maintain that header line accurately (and the TSV mirror's header, if you keep one).

Product backlog format

The skill writes simple entries to product-backlog.md (title, description, source, context, tag). If you already maintain a richer product features doc with priorities, an inspiration field, or a Pourquoi section, point the skill there by updating the path in SKILL.md and adjusting the entry format to match your existing structure.

Trigger phrases

The skill activates on phrases like "process this discovery call" or "run the master". If you prefer different trigger phrases, edit the description field in the skill's frontmatter — Claude reads that to decide when to invoke the skill.

---

A last note. The skill writes to your documents. It does not read them for you. Sit down with your insights.md once a week. Sort your CRM by ICP score and call back the top 5. Skim your competitor map before a sales call. Re-read your product backlog before a planning session. The skill keeps your docs current and structured; the actual steering of your startup happens when you read and process what you have captured.