Skill Build Kit — Production Workflow

How to Use This Skill

Follow this workflow in order. Complete the concept brief (01-context.md) before starting. Do not write the quick-start until the skill file is tested and producing correct output.

Step 1: Complete the Concept Brief

Open 01-skill-build-context.md and fill in every field.

Load before writing anything:

• business-aos/reference/core/voice.md — tone, cadence, vocabulary
• business-aos/reference/core/audience.md — who they are, their language
• 03-skill-build-golden-example.md — patterns to match

Validation gate: Do not proceed until:

• The skill's job is clearly defined (what goes in, what comes out)
• Output sections are listed with their purpose
• Trigger phrase is chosen
• Core principle is articulated

Step 2: Build the Skill File

Produces: [skill-name].md

Frontmatter

---
name: [skill-name]
description: >
  [Multi-line description. What it does, what input it takes,
  what output it produces, when to use it.
  Triggers: "trigger 1", "trigger 2", "trigger 3", or any request
  to [description of the job].]
metadata:
  author: Kathryn Brown, Advisory OS
  version: "[X.0.0]"
  updated: "[YYYY-MM-DD]"
---

The description field does double duty: it describes the skill for humans AND includes trigger phrases that Claude uses to recognize when to activate it. Write it conversationally — "Use when you want to prep for a client call, catch up after being heads-down, or hand off client context to a team member."

Heading + Positioning

# [Skill Name]

[1-2 sentences — what the skill gives you. Not what it does technically. What the user walks away with.]

Core Principle

## Core Principle

**[Rule name.]** [2-3 sentences: what the rule means in practice, what goes wrong without it.]

The core principle is the single most important operating instruction. For CIB it's "Read before writing" — absorb all input before producing output. Every skill has one. Find it by asking: "What's the one thing that, if Claude skips it, the output is garbage?"

What This Skill Does

## What This Skill Does

[Intro: what user pastes, what they get back. One paragraph.]

**Job 1: [Name]** — [Practical description of what this part of the output does for the user.]

**Job 2: [Name]** — [...]

**Job 3: [Name]** — [...]

Three jobs. Each job describes a user outcome, not a technical process. "Tells you what to say when the call starts" not "Analyzes communication patterns."

Section-by-Section Spec

## The [Output Name]: Section by Section

### 1. [Section Name]
[What this section contains. Why it matters. Format rules — be specific.]

### 2. [Section Name]
[...]

This is the longest part of the skill file. Every output section gets its own H3 with:

• What it contains — the information this section surfaces
• Why it matters — what the user does with this information (ties back to a job)
• Format rules — table structure, column names, bullet format, line count limits

Format rules must be specific enough that Claude produces consistent output across runs. "Two-column table. Numbers, Pattern, Completing, Deferring, Read — one row each." not "Present this clearly."

Rules

## Rules

- **[Bold label.]** [Explanation — what to do, why.]
- **[Bold label.]** [...]

Rules cover four categories:

1. Output format — no paragraphs, tables over prose, scannable, file naming
2. Scope — one input set per run, recency window, don't mix
3. Quality — don't assume, cite evidence, numbers over labels, edge case handling
4. Conventions — naming, [Rolling] tags, first-person language in usable sections

Cross-Referencing Other Skills (Series Skills Only)

If this skill is part of a series or accepts another skill's output as input:

## Reading a [Other Skill Full Name] as Input

[Section mapping: which sections of the other skill's output map to which
detection patterns or inputs in this skill. Table format preferred.]

Rules for cross-references:

• Always use the full skill name — never abbreviate ("Client Intelligence Brief skill" not "CIB")
• Remind the user the other skill exists ("if you don't have it, ask for it" or similar)
• If the skill can accept another skill's output OR raw input, both paths must be documented as equal
• The mapping section should cite the other skill's section labels as evidence sources

Output Format Template

## Output Format

[Complete markdown template with every section header, every table structure,
every label/value pair, every placeholder. Copy-pasteable.]

This is the blueprint. If someone printed this template and filled it in by hand, the result would match what the skill produces. Every header, every column name, every structural element must be present.

What Makes This Different

## What Makes This Different

[2-3 paragraphs. Speaks to the user, not to Claude.
Contrasts with the generic alternative.
Uses specific language from real testing results.]

This section earns the download. It answers: "Why should I use this instead of just asking Claude to summarize my emails?" The answer is always specificity — name what the skill catches that a generic prompt misses.

Step 3: Test the Skill

Before proceeding to the quick-start guide:

1. Start a new Claude conversation
2. Upload the skill file
3. Use the trigger phrase with real data (not fabricated input)
4. Compare output against the section-by-section spec
5. Check: Does every section appear? Does the format match? Are edge cases handled?
6. Fix and re-test until output consistently matches the spec

Minimum testing:

• 3 different inputs from real scenarios
• At least one minimal-input test (e.g., one email instead of five)
• At least one edge case (missing context, ambiguous data)

Step 4: Build the Quick-Start Guide

Produces: quick-start.md (or quick-start-[skill-name].md if multiple skills ship together)

Structure:

# [Skill Name] — Quick Start

You just got a free Claude skill. Here's how to use it.

## What You Need

- A Claude Pro account (or higher) — Skills require a paid plan
- [Input description — what they need to have ready]

## Setup (One Time)

### Step 1: Enable Code Execution

Skills require Code execution to be enabled. If you've never turned this on:

1. Click your name or initials in the bottom-left corner
2. Click **Settings**
3. Click **Capabilities**
4. Toggle on **Code execution and file creation**

If this is already on, skip to Step 2.

### Step 2: Upload the Skill

1. In the left sidebar, click the **Customize** icon (briefcase icon)
2. Click **Skills**
3. Click the **"+"** button
4. Select **"Upload a skill"**
5. Upload the ZIP file you received
6. Toggle the skill **on**

That's it. The skill is installed.

## Running the Skill

1. Start a new conversation in Claude
2. Type: **"[Trigger phrase]"**
3. Paste [input description]
4. Wait about [time estimate]

[1-2 sentences describing what they'll get back — the output, not the process.]

## Tips

**[Tip from testing.]**
[One sentence explanation.]

**[Tip from testing.]**
[One sentence explanation.]

**[Tip from testing.]**
[One sentence explanation.]

## What to Paste

Any of these work:

- [Input type 1]
- [Input type 2]
- [Input type 3]
- [Input type 4]

[One sentence about what the skill handles.]

---
*Built by Kathryn Brown — Advisory OS*

Rules for the quick-start:

• Every instruction references exact UI elements ("Click your name in the bottom-left corner" not "Navigate to settings")
• Tips come from real testing, not generic advice
• "What to Paste" lists specific input types, not categories
• No marketing language — this is a setup document

Step 5: QC Both Assets

Run 04-skill-build-quality.md against both files.

Blocking checks:

• Skill file YAML frontmatter complete
• Output format template is copy-pasteable
• Tested on real data with consistent output
• Quick-start has exact UI paths
• Voice check passes — locked terms used, no forbidden constructions

Fix all failures before shipping.

Step 6: Deliver

Final files:

• [##]-[skill-name]-skill.md — the skill file (numbered by series position)
• [##]-[skill-name]-quick-start.md — the quick-start guide (same number)

Both go to the campaign folder (e.g., content/business/marketing/campaigns/[campaign-name]/).

Naming convention: All campaign assets use [##]-[skill-name]-[asset-type].[ext]. The ## is the skill's position in the series (01, 02, etc.). Asset types: skill, quick-start, delivery-page, email, dm, post.