Brand Context Builder

What it does

Every skill in this catalog reads three files at the project root before it does anything: BRAND.md, ICP.md and VOICE.md. This skill runs the interview that produces them. It is the highest leverage twenty minutes in the whole catalog, because after it every other skill starts from a warm start instead of a questionnaire.

When to use it

• set up my context
• the skills keep asking me the same things
• start a new engagement
• write my brand and ICP files
• bootstrap

What it needs from you

• business: what you sell, to whom, and what it replaces
• customers: your best customers and why they were good
• writing: two or three things you actually wrote

How it works

1. Ask the smallest set of questions that produces all three files. Do not ask anything you can read from the website, the existing files, or the samples.
2. Write BRAND.md: what the business does in one sentence a customer would recognize, the proof points with numbers, the positioning against the two real alternatives, the words to use and the words never to use, and the claims that are not allowed because they cannot be supported.
3. Write ICP.md: the firmographic filters as ranges and lists, the signals, the pain indicators, the exclusions, and the tier rule. If the ICP definition skill has run, this is its output rather than a second attempt at it.
4. Write VOICE.md: sentence length, contraction use, how the person opens and closes, the three phrases they actually use, and the register, derived from the writing samples rather than described in the abstract.
5. Mark every unknown as unknown rather than filling it with something plausible. A confident wrong proof point propagates into every email the catalog writes.
6. Set a review date on each file and record the date it was written, because positioning drifts and a stale voice file is worse than none.
7. Tell the user which skills change behavior now that the files exist, so the value is visible immediately.

What you get back

Three files written at the project root, each with a written date and a review date, plus a short list of what is still unknown and which skills are now warmer.

The quality bar

• All three files are written, or the missing ones are explained
• Unknowns are marked as unknown rather than filled in
• Voice markers are derived from real samples, not described in the abstract
• Every file carries a written date and a review date
• Disallowed claims are recorded in BRAND.md

Where it runs

• Claude Code
• Codex CLI
• Cursor
• MCP-compatible agents