Claude Gtm Skills

claude-gtm-skills

A production library of 10 Claude Agent Skills that enforce outbound and content standards at generation time.

These are not prompt templates. Each skill is a specification with hard gates: required inputs that must be supplied before generation begins, banned-token lists checked against every draft, and explicit pass/fail or numeric scoring that runs before output is delivered. The model is not asked to write well — it is constrained until it does, or it stops and asks.

Built and used daily across client engagements in B2B tech, real estate, procurement, and professional services.

---

Why this exists

Cold outbound written by an LLM fails in predictable ways. It invents proof. It reaches for `leverage`, `seamless`, `unlock`. It frames pain as the reader's fault. It closes with "would love to connect." It hedges. Every one of those failures is detectable, so every one of them is preventable at the spec level rather than the review level.

The library encodes that. A skill refuses to produce copy when it lacks the inputs to produce it honestly — if no real proof has been supplied, it asks rather than fabricating a number. That refusal behavior is the point.

---

Architecture

Ten skills across four layers. Routing is handled in the description field of each skill, including negative routing that names its sibling skills to prevent the wrong one firing.

Cold email

Skill	Register	Enforces
kevin-oleary-sales-email	Tight, high-volume	11 non-negotiables, pain-first ordering, binary close, 8/10 minimum score before delivery
casual-peer-proof-sales-email	Warm, slower-burn	Peer proof stated as third-party fact, no manufactured urgency, anti-AI-slop token list

These two deliberately overlap and explicitly route away from each other. The O'Leary skill handles volume outbound with a one-sentence constraint and a binary close. The casual skill handles trust-first plays where two-to-four sentences an