Brand Context
Build and maintain one brand-context.md file per brand: the single source of truth for what the business sells, who it sells to, how it talks, and what it can prove. Every other skill in this collection reads this file before asking the user brand questions, so the user explains their business once — not once per task.
This is the hub in a hub-and-spoke model. Ad copy, blog posts, cold email, social content, and landing-page work are all downstream of the same positioning and voice; when those live in one maintained doc, every downstream skill starts warm and stays consistent.
The file contract. The doc lives in persistent storage at a stable path —
/files/<brand-slug>/brand-context.md— and is reused at that SAME path every time. Create it once, then update it in place; never recreate it from scratch when it already exists (you would destroy verbatim customer language and proof points the user curated). If a run touches the doc, bump theLast updatedline.
Requirements
- Hyper MCP connected. https://app.hyperfx.ai/mcp — needed for the file tools that persist the doc, and for auto-draft.
- Recommended for auto-draft, enabled at https://app.hyperfx.ai/apps: Firecrawl (read the brand's site and extract visual identity). Without it, skip auto-draft and use the interview path — the skill still works.
- Optional enrichment: the Apify scrapers used by
customer-research(verbatim customer language) and HyperSEO (competitive landscape). Both are nice-to-have; mark any section built without real data as unverified rather than inventing content.
How to run the tools in this skill
Every tool in this skill is named by its canonical tool name. Run it with the call your surface gives you:
| Surface | Find a tool | Run it |
|---|---|---|
| Hyper chat | Call the tool by its name; search_tools(queries=["<provider> <action>"]) shows its parameters | <name>(...) |
| MCP client (Claude, Cursor, Codex, ChatGPT) | search("<what you want to do>"), then describe("<name>") | call_read("<name>", {...}), call_write or call_destructive: the one that describe names in call_with |
| Hyper CLI | hyperai search "<what you want to do>", then hyperai describe <name> | hyperai call <name> --json '{...}' |
If a tool is not found: in Hyper chat, find its toolkit id with discover_toolkits and pass it to enable_toolkit. On an MCP client or the CLI, tell the user which integration to connect in Hyper.
Tool surface
| Job | Tools |
|---|---|
Read, create, update brand-context.md | your file tools (read_file, create_file, edit_file) |
| Auto-draft from the brand's site | firecrawl_urls_scrape (homepage, about, pricing, product pages), web_pages_scrape fallback |
| Visual identity (colors, logo, typography) | firecrawl_branding_extract |
| Verbatim customer language (optional) | defer to customer-research |
| Competitive landscape (optional) | defer to competitor-intel, seo-research |
Out of scope: defer to other skills
| Request | Send them to |
|---|---|
| Deep voice-of-customer mining (Reddit, reviews, social) | customer-research |
| Full competitor research, battle cards, monitoring | competitor-intel |
| Blog topic strategy and the published-post log | blog-generation (its blog-strategy.md Brand brief should be seeded from this doc, not rebuilt) |
| Extracting brand assets for image generation | ad-creative-generation (firecrawl_branding_extract at creative time) |
| Keyword or SERP research | seo-research |
The document
The full template with per-section guidance lives in references/brand-context-template.md. The sections, in order:
- Product overview — what it is, what it does, category, one-liner.
- Target audience / ICP — the specific customer, firmographics or demographics, who it is NOT for.
- Personas — 2–4 buyer/user personas with role, trigger, and success criteria.
- Pain points — the problems that drive purchase, ranked.
- Competitive landscape — direct and indirect alternatives, including "do nothing" / spreadsheets / status quo.
- Differentiation — why customers pick this brand over each alternative; what only this brand can claim.
- Objections & switching — what stalls deals, what prospects fear, what switching costs them.
- Customer language — verbatim quotes: how customers describe the problem and the product in their own words. The most valuable section in the doc.
- Brand voice — tone, reading level, point of view, words to use, words to ban.
- Proof points — real numbers, named customers, credentials, data. Only verified items belong here.
- Goals & current focus — what marketing is trying to move this quarter, priority channels.
- Assets & channels — site URL, blog, social handles, ad accounts in play, visual identity notes.
Building the doc
Step 0: check for an existing doc. Look for brand-context.md at the brand's path (and reasonable legacy locations). If one exists, this is an update run: read it fully, then only revise the sections the user wants changed or that new evidence contradicts. Only when nothing exists do you build fresh.
Path A — auto-draft (default when the brand has a website).
- Ask for the site URL if you don't have it.
- Scrape the load-bearing pages with
firecrawl_urls_scrape: homepage, about, pricing, 1–2 product/feature pages, and a customer-facing page with testimonials or case studies if one exists. - Optionally call
firecrawl_branding_extractonce to fill the visual identity notes in section 12. - Draft every section you have real evidence for. Where the site is silent (objections, goals, most customer language), leave the section marked
*(unverified — needs input)*instead of guessing. - Present the draft to the user section by section for correction before saving. The site says what the brand wants to be true; the user knows what is true. Ask specifically about: the ICP ("who actually buys?"), differentiation ("why do people pick you over X?"), and proof points ("which of these numbers are real and current?").
- Save to
/files/<brand-slug>/brand-context.mdand report the path.
Path B — interview (no site, pre-launch, or user prefers talking).
Walk the sections conversationally, a few questions at a time, in the order given in references/brand-context-template.md — not as one giant questionnaire. Start with product overview and ICP (everything else hangs off those), and let the user skip sections; an honest partial doc beats a padded complete one. Save incrementally so a broken-off interview still leaves a usable doc.
Either path, before saving: run the quality gate below.
Quality gate
- No fabrication, anywhere. Never invent a customer quote, a stat, a competitor claim, or a persona detail. An empty section marked "needs input" is correct; a plausible invented one is corrosive — downstream skills will confidently repeat it in ads and emails.
- Verbatim means verbatim. Customer language is quoted exactly as customers said or wrote it, with a source (review, call, Reddit thread). Polished paraphrases go in positioning, not in section 8.
- Specific beats complete. "Ops managers at 20–200-person e-commerce brands drowning in carrier spreadsheets" is useful; "businesses that want to save time" is filler. Cut filler.
- Proof points are verified. Only numbers and names the user confirmed. Date them ("2,400 customers as of Jun 2026").
- The one-liner passes the stranger test. Category + audience + core value in one sentence a stranger could repeat.
How other skills consume this doc (the spoke contract)
When any marketing skill needs brand context — positioning, audience, tone, proof:
- Check for
brand-context.mdfirst at/files/<brand-slug>/brand-context.mdbefore asking the user a single brand question. If it exists, read the whole file and ask only for what the task needs that the doc doesn't cover. - Honor the voice section — the words-to-use / words-to-ban lists are hard constraints on generated copy.
- Pull hooks from customer language — verbatim phrases from section 8 outperform invented copy; prefer them for headlines, subject lines, and openers.
- Never cite a proof point that isn't in section 10. If the doc has no proof for a claim, the copy doesn't make the claim.
- Check freshness. If
Last updatedis more than ~6 months old, flag it and offer a refresh — but do not silently rebuild the doc mid-task. - Write back nothing. Spokes read; only this skill (with the user) writes. If a task surfaces something that belongs in the doc — a great customer quote, a new competitor — suggest the user run
brand-contextto add it.
Maintaining the doc
The doc is alive. Update it when positioning shifts, a new competitor matters, pricing changes, or research (from customer-research / competitor-intel) yields better quotes and landscape data. On update runs: read first, edit only what changed, keep everything the user curated, bump Last updated, and summarize the diff back to the user. For multiple brands, keep one doc per brand under its own <brand-slug> directory — never mix brands in one file.
References
references/brand-context-template.md: the full copy-paste document template, per-section guidance on what good looks like, and the interview question bank for Path B.
