Starting a New Engagement

What You Need Before Starting

Before running the scaffolding, three things from the client:

A brief: Can be a raw discovery doc, a Slack thread, a one-pager, or a voice memo transcript. Marketing Bootstrap reads it and structures it. The rougher the input, the more you’ll need to review the output, but the tool handles unstructured input.

Brand materials: Logo files, existing copy (website, decks), any style guide. These become the inputs for voice.md and style-guide.md. If none exist, we write them from scratch in the first week.

Notion access: An empty Notion workspace where the databases will be created.

What the Scaffolding Produces

Running the init command creates:

<client-name>/
├── CLAUDE.md                    # Claude Code project context
├── core-md/
│   ├── voice.md                 # Tone, banned phrases, stylistic anchors
│   ├── positioning.md           # Core value props, competitive differentiators
│   ├── product-facts.md         # Features, specs, technical details
│   ├── audience-personas.md     # ICP definitions with pain points
│   └── style-guide.md           # Visual identity rules
├── agents/                      # Agent configuration files
├── skills/                      # Wrapper command implementations
└── outputs/                     # Where generated content lands

The core-md/ files are drafted from the brief. They’re starting points. You review and approve before any agents run against them.

First Commands to Run

Once the environment is set up and core-md files are approved:

1. /weekly-report: Generates the baseline report template. Low risk. Shows the system working end-to-end before any content production.

2. /write-blog <intro-topic>: First content piece. Use a topic that doesn’t require competitor data so the full pipeline runs without gaps.

3. /seo-pulse: First market scan. Shows the analyst agent’s output before any campaign work begins.

Reviewing Core-MD Quality

Everything Clembot produces depends on the quality of the core-md files. Before any client-facing output, review:

voice.md: Read it out loud. Does it sound like the client? Would someone mistake an AI-written paragraph for theirs? If not, edit until they would.

positioning.md: Are the differentiators specific? “Easy to use” is not a differentiator. “Only RPC provider with sub-50ms latency on Base” is. Vague positioning produces generic content.

product-facts.md: Is every claim verifiable? The editor agent checks against this file. Weak facts produce low compliance scores and content that has to be rewritten.

Weak core-md files are the most common reason a first engagement produces mediocre output. The setup investment pays back immediately.