Brand DNA Interview . Agency OS

What it does

Produces three files that every other Agency OS skill reads: owner-profile.md (your story, mission, methodology, voice, sales intelligence and goals), epistemology.md (how you actually think and diagnose problems), and policy.md (what the AI may never touch). It also creates verified-claims.md, the registry of your real, consented proof that downstream copy is allowed to draw from. Before the interview it scrapes your website and recent social posts to extract your existing voice patterns, and it can mine call transcripts for the way you really talk.

Say this to start

This skill has no button. You start it by saying what you want. Any of these will do it:

/build-profile

/first-win

/deepen-profile

> brand DNA interview

> build my owner profile

> capture my brand voice

When to reach for it

First build in a brand new workspace
Onboarding a new client and capturing their brand
Output keeps sounding generic or not like you
Profile is more than 30 days old
Upgrading a lite profile to a full one
You have call recordings you want mined for voice

When NOT to use it

If you actually want	Use this instead
Profiling the customer the business sells to	`ideal-customer-avatar`
Packaging and pricing what the business sells	`offer-architect`
Writing copy in the voice once it is captured	`legendary-copywriters`
Visual and art direction for the brand	`creative-director`
Adding only the epistemology layer to a profile you already have	`/deepen-profile`

Before you start

What you need	Why
About 15 minutes of uninterrupted attention	the interview is confirmation-gated and probes when answers are vague, so a rushed pass produces a thin profile	Required
Your website URL and social handles	voice extraction scrapes the site and your last 20 to 30 posts before the interview. Without them you answer two extra voice questions instead	Optional
A web scraping tool (WebFetch, Firecrawl or Playwright)	no tool means no scraping, and you paste two or three writing samples instead	Optional
Real proof you can document: named clients, case studies, screenshots, public links	`verified-claims.md` only accepts what you attest to. An empty registry is allowed, but every downstream output then stays generic and flags a proof gap	Optional
Call recordings or transcripts you have the right to share	conversation mining produces a far deeper epistemology than the interview alone. Without them the file is flagged as a shallow source	Optional
Your answer to who speaks for the brand: the company, one named person, or both	this single answer decides whether every downstream skill writes we or I	Required

How it runs

Pre-interview checkLooks for an existing owner-profile.md in your workspace. If it is there it offers to update, deepen, or start fresh, and shows a diff before overwriting anything. Existing context is pre-loaded so you are not asked the same things twice.
Voice extraction (before any questions)Asks whose voice this is (company, personal, or both), your exact posting name, industry, website and social handles. Then it scrapes and analyses your existing content and shows you the patterns it found, waiting for you to confirm they are accurate.
Optional conversation miningOffers to ingest call recordings or transcripts. You can paste one straight into the chat and it will ask before filing it into raw-conversations/. It scans first, shows you a cost estimate, and waits for your go-ahead before spending anything. Default cap is three transcripts.
Sections 1 to 3Origin story and the specific turning point, mission and values and your hard no's, then your methodology and named frameworks. It probes with follow-ups when an answer is generic.
Section 4: the way you see thingsSix questions, five to seven minutes, described in the source as the highest-leverage section in the interview. Assumptions you keep correcting, your reframes, your diagnostic sequence, what you trust as evidence, questions you ask repeatedly, and distinctions you draw. This section is blocked from advancing until it has enough to meet the epistemology minimums.
Sections 5 to 9Voice and signature phrases, content platforms and topics, how you actually win clients and handle objections, quarterly and 12-month goals, and finally the boundary list of what the AI must never touch on its own.
SynthesisCompiles owner-profile.md, then epistemology.md with a confidence tag on every pattern, then policy.md. Shows you the complete profile and waits for corrections before saving.
CloseoutRuns the quality gate, saves the .docx, uploads to Google Drive if gws is installed, updates and opens the dashboard, asks whether you want slides, and logs how the session went to learnings.md.

What you get

brand/[workspace]/owner-profile.md, the canonical profile every skill reads
brand/[workspace]/epistemology.md, how you think, with confidence tags and concrete examples per pattern
brand/[workspace]/policy.md, the publishing rule plus your never-touch and heightened-review lists
brand/[workspace]/verified-claims.md, your real proof registry with consent flags
brand/[workspace]/brand-dna/brand-dna-output.md, the consolidated dossier (the build is incomplete without it)
brand/[workspace]/brand-dna/conversation-extracts.md when transcript mining runs
brand/[workspace]/voice-profile.md and the matching .docx files
A dashboard section and an optional slide deck

Honest limits

Read this before you rely on it

The epistemology section is blocking. If your answers stay abstract it will re-probe rather than move on, and the build does not finish without at least three invisible assumptions, three reframes, one diagnostic sequence and three recurring questions.
Conversation mining costs money. It spawns one sub-agent per transcript, which is why it shows you an estimate and refuses to start until you approve.
Only text transcripts can be pasted or attached. Audio and video files have to be dropped into raw-conversations/ yourself so they can be transcribed locally.
With no website, no social and no transcripts, the voice profile is built from interview answers alone and the epistemology file is flagged as a shallow source. The lite path (/first-win) skips voice extraction, scraping, mining and epistemology entirely: it unblocks the other builds, it does not replace this one.
Only proof you can actually document belongs in verified-claims.md. If you leave it empty, every downstream skill writes generic copy with a proof-gap marker, which is the intended behaviour, not a fault.

Where people go wrong

The mistake	Do this instead
Skipping the interview and going straight to the ICA or the offer	Every downstream skill reads this profile. Run it once, or run `/first-win` for the lite version, and the rest of the system stops sounding generic.
Giving the vague version of the origin story	Name the specific day, the specific client, the specific failure. Contrast and dates are what make the story usable in copy.
Inflating proof points to sound more impressive	Log only what you can show: a named client, a written case study, a screenshot, a public link. Unverifiable numbers collapse the first time a buyer asks who says so.
Answering the voice section with the style you think you should have	Answer with how you actually talk to clients, including the phrases you overuse and the words that make you cringe.
Assuming the profile will be re-run wholesale to fix it	The deep layer fills in progressively over later sessions, and `/deepen-profile` adds the epistemology layer to an older profile without redoing the interview.

Worth knowing

Real call recordings beat polished published content. The richest voice and thinking signal comes from unfiltered conversation, so paste a sales or coaching call transcript into the chat when it offers to mine one. Interview answers alone get the file flagged as a shallow source.