branddock-on-brand

Work on-brand with Branddock, the brand memory and brand check for AI. Use this skill whenever you create, rewrite or review content for a brand that uses Branddock (social posts, emails, blogs, ads, web copy, scripts) — it makes you fetch the real brand context first, write with it, score every out

Branddock — brand memory and brand check

Branddock holds a brand's complete DNA (voice, personas, products, competitors, guidelines) and can measure how on-brand any text is: the F-VAL score (0–100, with concrete findings). Your job with this skill: never write "generic AI" content for a Branddock brand. Fetch the brand first, write with it, verify with the score, and let the human approve.

When to use this skill

• Any content task for a brand that has a Branddock workspace: writing, rewriting, reviewing, translating, brainstorming in brand voice.
• Any question about the brand itself ("how do we sound?", "who are our personas?", "what does our brand say about X?").
• Checking whether an existing text (from any source) fits the brand.

Not needed for: content explicitly unrelated to the brand, or pure technical/code tasks.

How you connect (pick the first that is available)

1. MCP connector (preferred): tools from the Branddock connector (https://branddock.app/mcp). If the tools are listed in your session, use them directly.
2. REST API (for scripts/automation): base https://branddock.app/api/v1, header Authorization: Bearer bd_live_... (workspace-locked API key, created in Branddock under Settings → API & Connectors).
3. Neither available? Tell the user: add the connector URL https://branddock.app/mcp in your AI's connector settings and log in with your Branddock account — or create an API key in Settings → API & Connectors. Don't improvise brand knowledge without a connection.

The workflow (always this order)

1. Know the brand — before writing anything

• Call get_brand_context once per conversation (per brand). It returns the full brand DNA: voice and tone rules, do's & don'ts, personas, products, competitors, style notes and the brand's content language.
• Multiple brands on one account? Call list_brands first and pass the right brand (workspace id or exact name) to every tool call.
• Ground every brand statement in this context. If the context doesn't answer something, say so — never invent brand facts, slogans, product claims or persona details.

2. Create with the brand, not next to it

• Write in the brand's content language (it is in the context), unless the user explicitly asks otherwise.
• Apply the voice rules literally: vocabulary to use and to avoid, tone dimensions, anti-patterns. Personas from the context are your audience definitions — pick the one the user names, or ask.
• For substantial deliverables (posts, blogs, campaigns) prefer generating through Branddock (generate_on_brand) over writing inline: the full brand pipeline runs server-side, the result lands scored in the user's Branddock library, and the user keeps one place to review. Inline writing is fine for quick drafts and rewrites — then score it (step 3).
• Rewriting an existing text (email, reply, external copy)? Use rewrite_on_brand — nothing is stored, costs 1 credit.

3. Verify — score before you present

• Score your result with score_against_brand (free, unlimited). You get a composite 0–100, a per-pillar breakdown, the brand's threshold and concrete findings.
• Below the threshold? Revise using the findings (they tell you what is off-brand) and rescore. Maximum two revision rounds — if it still scores low, present your best version with the score and findings, honestly. Never hide a low score.
• Mention the score when you deliver: "F-VAL 84, above the threshold" is part of the answer, not decoration.

4. The human approves — always

• You never publish, schedule or send anything yourself. Everything you make is a proposal.
• Content generated through Branddock waits in the user's library for review; inline drafts are handed to the user with their score. Say what the next step is ("review it in Branddock" / "want me to adjust?").

Tool map (MCP names)

Know the brand (free): get_brand_context · get_brand_md · list_brands · list_personas · list_products · list_competitors · search_knowledge · get_deliverable_content Guard & follow (free): score_against_brand · get_strategy_status · get_seo_status Make (costs credits — see below): generate_on_brand · rewrite_on_brand (1) · generate_web_page (5) · generate_long_form_seo (80) · generate_campaign_strategy (80)

API-key only — NOT available over the MCP connector: generate_image (2) · generate_video (20) · import_brand_data. Over the connector (route 1 above) these tools do not exist; don't try to call them. Need on-brand imagery or video? Tell the user it runs through a Branddock API key (route 2) or in the Branddock app itself.

REST equivalents (when using an API key)

bash

# Full brand context
curl https://branddock.app/api/v1/brand-context \
  -H "Authorization: Bearer bd_live_..."

# Score a text against the brand
curl https://branddock.app/api/v1/score \
  -H "Authorization: Bearer bd_live_..." \
  -H "Content-Type: application/json" \
  -d '{"content": "... (min. 50 tekens) ..."}'

Generation endpoints mirror the MCP tools (/generate, /seo-generate, /webpage-generate, /image-generate, /video-generate).

Credits, costs and honesty

• Reading brand context and scoring are always free — use them liberally; that is the point of Branddock.
• Generating costs credits (numbers above). Before starting an expensive generation (long-form SEO or a campaign strategy, 80 credits; video, 20), tell the user the cost and expected duration (SEO ≈ 7–8 minutes, async — poll with get_seo_status).
• Never claim autonomy the product doesn't have: Branddock's own rule is "you approve" — no "autopilot" language, ever.

When something fails

• 401 / unauthorized: the connector session or API key is invalid — ask the user to reconnect (connector) or check the key.
• Credits exhausted: say so plainly; the user can top up in Branddock (Settings → Billing). Free tools keep working.
• Brand not found: call list_brands and use an id or exact name from that list.
• A tool call needs the user's permission (chat clients ask per tool): explain that read/score tools are safe to always-allow; generation tools are worth approving per use.

First prompts that work well

• "Haal mijn merkcontext op en vat samen hoe mijn merk klinkt."
• "Schrijf een LinkedIn-post over [onderwerp] voor persona [naam] en scoor hem met de merk-check."
• "Herschrijf deze e-mail in onze merkstem en laat de score zien: [tekst]"