Inventory Your Harness: The Six Components

Part 4 of the field-guide companion to Builder-Leader. The first three parts diagnosed where you stand. This one opens the exoskeleton and names what it's made of, because the rest of the ladder assumes you can point at any component and say "mine's weak here."

The six components

A harness is the structure you build around a model. It has six parts:

• Skills: reusable instructions for a recurring task. At its simplest, a markdown file that tells the system how to do the thing you keep asking it to do.
• Memory: persistent context the system reads every session. A CLAUDE.md with your stakeholders, your active workstreams, your tone.
• Agents: scoped helpers you can hand a bounded task and let run.
• Hooks: deterministic rules that fire on events, the guardrails that stop a class of mistake.
• MCP: Model Context Protocol servers that connect the system to your tools and data.
• Sensors and gates: the checks that decide whether output is good enough to pass, and the signals that tell you when something drifted.

That's the vocabulary for everything that follows. Learn to name the six and you can describe any setup, yours or a peer's, precisely.side quest

Inventory what you actually have

Draw six rows, one per component. For each, write three things:

• Have: what's in place today, even informally. A recurring prompt you paste counts as a proto-skill. A notes file you share with the model counts as proto-memory.
• Use: whether you actually use it weekly. Honest answer.
• Gap: one sentence on what's missing, or why "have" and "use" don't match.

Most executives doing this honestly find zeros across the board: no built skills, no persistent memory, no agents, no hooks, no MCP, no gates. Not because they're lazy. Because nobody ever showed them these are named things with specific tools behind them. The inventory turns "I should use AI more" into a concrete list of six addressable gaps.

Turn the moat argument into a renewal calendar

The book opens Chapter 4 with a commodity-versus-moat argument: the model is a rental everyone can get, and a growing pile of standalone SaaS tools are being absorbed by a model plus a harness plus a skill or two. Claude Design producing usable design work directly is the clean example. The point isn't any single tool; it's the pattern, and you can find it in your own spend.

List the top ten SaaS tools your department pays for. For each, answer one line: could a harness plus a model plus one or two skills replace this within 18 months? Three buckets fall out:

• Definitely replaceable: reporting dashboards a skill can generate, drafting and templating tools, design assets a model produces directly.
• Probably not: anything tied to regulated pipelines, specialized hardware, or external network effects.
• Unclear: the interesting column. Sit with it.

Hand the "unclear" column to a procurement review and ask whether any of those renewals are worth delaying a quarter while you run a harness pilot. The moat argument stops being a thesis the moment it's a line on your renewal calendar.side quest

Pick your first build

Go back to the six-row inventory. Find the row where the gap is largest and the build is simplest. For almost everyone that's Skills (a markdown file with instructions) or Memory (a CLAUDE.md the system reads every session).

Name the artifact in one sentence. Not "I'll build a skill" but: "a skill called weekly-status-draft that knows the five sections of my status note and the three things my manager always asks about." Or: "a CLAUDE.md with my stakeholder list, three active workstreams, and the tone I want in drafts."

That named artifact is your week-one target when you start the hands-on ladder (Part 6 of this series). Pre-picking it is how week one avoids stalling on "what do I build first."

The full Chapter 4 field guide is in the public repo: field-guide/04-harness-and-components.md. The book is at builder-leader.com.

Next: the five leadership primitives that already transfer to a harness, and the one that needs reframing.