Not Every Skill in the Pack Was Built Here

When we shipped the Freedom Ag skills module, five zip files went out. We didn't build all five of them.

humanizer is open source, MIT license, from GitHub. pdf is an Anthropic built-in that ships with the Claude client. spin-selling, influence, and storybrand are working summaries we wrote: frameworks from published books, not copies of the books.

Three sources. That split was deliberate.

Why Not Build Everything

Building a skill requires articulating a framework clearly enough that Claude applies it consistently. That's the slow part. If someone has already done that work and published it under an open license, building a parallel version is wasted effort.

humanizer is a good example. The open-source version strips the tells that make writing read as AI-generated: the inflated transitions, the three-item lists, the constructions AI defaults to. The framework is well-defined and the implementation is solid. We tested it against the ag emails the client was already writing and it performed. There was nothing to gain by rebuilding it.

pdf is a different case. It's an Anthropic built-in, meaning it ships with the Claude client without any installation. Including it in the pack wasn't about adding a capability that didn't exist. It was about pointing the client to a capability they already had and hadn't used. The skill appears in their list now. That's enough.

Where Original Work Comes In

The three book summaries didn't exist as ready-to-download skills for an ag inputs sales rep. General versions of these frameworks are everywhere, but general doesn't fit.

A general SPIN selling skill teaches the four question types. An ag-specific one tells you when each type lands with a grower who's been buying from the same supplier for 15 years and won't change on urgency. That context lives in the skill file because we put it there.

Building these from scratch also meant adding guardrails that matched the client's actual risk. The influence skill flags manufactured scarcity and authority explicitly, because those are the two most abused tactics in ag supply sales and the client didn't want to use them. A generic Cialdini skill wouldn't have that constraint. We needed to write it in.

What the README Says About It

We covered the sourcing in one paragraph:

humanizer is open source (MIT). pdf is one of Anthropic's built-ins. spin-selling, influence, and storybrand are working summaries of the books they're named after, they're commentary and frameworks, not copies of the books. If one of them earns its keep, buy the book.

The last sentence matters. A skill summary is not the same as reading the book. If spin-selling changes how someone runs a discovery call, the actual Rackham research will return more than the skill did. Pointing to that directly felt more honest than implying the skill covered everything.

The Pattern Going Forward

When we build a skills pack now, the first question is what already exists. Anthropic's built-in library covers infrastructure: pdf handling, morning digests, MCP server builders. The open-source ecosystem covers writing style tools, format utilities, and a growing range of domain-specific prompting patterns. A lot of what clients need is already there.

The original work goes where context can't exist in a general tool: the client's specific risk rules, the constraints their industry requires, the framing their customers respond to. That's the part no one else has already built.

A pack built entirely from scratch takes longer than it needs to and isn't necessarily better. A pack built entirely from existing tools fits no one specifically. The right split varies by client, but the question is always worth asking before you open a blank file.