Check If A Document Adheres To Our Brand Guidelines

Most brand breaches don’t come from bad intent. They come from a founder shipping a deck at midnight, a new hire guessing at the tone, or a contractor who never read the brand book. By the time someone notices, the file is already with the customer.

This tutorial builds a brand reviewer with HybridClaw. You hand it a document — a proposal, a deck, a landing page copy, a LinkedIn post — and it returns a prioritized list of fixes against your brand guidelines, not a generic checklist. Used before sending, it catches drift in minutes.

What We’re Building

1. You save your brand guidelines once as a single markdown file the reviewer can read every time.
2. You hand HybridClaw any document (.docx, .pptx, .pdf, .md, .txt, or a pasted URL or body).
3. HybridClaw scores it against the guidelines across five dimensions: voice, terminology, visual/typographic rules (where observable), messaging architecture, and compliance boilerplate.
4. It returns a prioritized list of fixes — must-fix, should-fix, nice-to-have — with the exact location in the document and a proposed rewrite where a rewrite is cheap.

This is a strong fit for agencies reviewing contractor output, SaaS companies keeping a consistent voice across sales and marketing, and founder-led teams where every new hire is one slide away from accidentally rebranding the company.

Prerequisites

Before starting, make sure you have:

• HybridClaw installed and the gateway running
• the bundled document tools available to read .docx, .pptx, and .pdf inputs; see Office Dependencies
• a brand guidelines document, even a rough one — one page is fine to start, you will improve it as the reviewer surfaces gaps

Step 1: Capture Your Brand Guidelines As One File

The reviewer is only as strict as the guidelines you give it. Do not paste a 60-page brand book on day one. Start with the rules you would actually enforce on a Monday morning.

Save the following as brand-guidelines.md in a known folder. Fill in each section with the shortest possible rules — bullets beat paragraphs.

# Brand Guidelines — <Company>

## Voice
- we sound like: <three adjectives, e.g. calm, precise, plainspoken>
- we do NOT sound like: <three, e.g. hypey, jargon-heavy, corporate>
- point of view: first person plural ("we") in marketing, first person
  singular ("I") in founder posts, never "the team" or "the company"

## Terminology
- always: <preferred terms, e.g. "customer" not "user", "platform" not
  "software", "data" as singular>
- never: <banned terms, e.g. "solution", "leverage", "utilize",
  "cutting-edge", "revolutionary", "synergy">
- product names: <exact casing, e.g. HybridClaw (one word, capital H, C)>
- always explain before acronym: first use "retrieval-augmented generation
  (RAG)", thereafter RAG

## Messaging Architecture
- one-line pitch: <the canonical sentence>
- three proof points: <the three claims we always back>
- audience priority: <e.g. operators first, developers second,
  executives third>
- what we do NOT claim: <e.g. no "AI that thinks", no "fully autonomous",
  no 10x productivity claims without a source>

## Visual & Typographic Rules (for slides and documents)
- primary typeface: <e.g. Inter for body, Söhne for display>
- color palette: <list hex codes, e.g. brand blue #1756E5, ink #0B0F19>
- logo: minimum clear space equal to the cap height of the logotype
- never: stretched, rotated, gradient-overlaid, or stroked logos
- screenshots: always on a neutral background, never on a product screen
  with live customer data

## Compliance & Legal Boilerplate
- every public-facing document that quotes a customer needs attribution
  and date
- never state an uptime or performance number without a linked source
- GDPR: never collect email without an explicit opt-in sentence on the
  same visible frame

## Non-Goals
- we are not a brand that preaches ("revolutionizing X")
- we are not a brand that apologizes ("humbly believe")
- we do not use emojis in body copy; they are allowed in social and
  slack posts only

A one-page version is enough to start. The reviewer will tell you which sections are vague; tighten them as you go.

Step 2: Test The Reviewer Manually

Open a local session:

hybridclaw tui

Upload or paste the document under review along with the guidelines file. Then run this prompt. The format is deliberately opinionated: must-fix first, exact locations, proposed rewrites where they are cheap.

🎯 Try it yourself

You are my brand reviewer. Read the attached brand guidelines, then
read the attached document, then return a structured brand review.

Guidelines file: brand-guidelines.md
Document under review: <filename or pasted content>
Document type: <proposal | deck | landing-page copy | social post |
               email | case study | press release>
Audience for the document: <e.g. mid-market CFO, developer advocate,
                             existing customer>

Review the document against these five dimensions, in this order:

1. Voice — does it match the "sounds like" / "not like" rules?
2. Terminology — any banned terms, any missing preferred terms, any
   product-name casing errors, any unexplained acronyms?
3. Messaging architecture — does it lead with our one-line pitch and
   back the claim with our proof points? Does it avoid the "do not
   claim" list?
4. Visual & typographic rules — only flag what is observable in the
   file (typeface in a .docx/.pptx, logo usage, color codes if present).
   Do not hallucinate; say "not observable" if you cannot see it.
5. Compliance — attributions, date stamps, performance claims with
   sources, GDPR opt-in where relevant.

Output format:

## Verdict
One sentence: ship-ready, ship with fixes, or do-not-ship, and why.

## Must-fix (blocks shipping)
Each item:
- **Location** — exact quote or page/slide/paragraph reference
- **Rule broken** — which guideline section and rule
- **Why it matters** — one sentence
- **Proposed rewrite** — if the fix is a phrase or sentence, propose
  it verbatim. If structural, describe the change.

## Should-fix (before next revision)
Same four-field format.

## Nice-to-have (style polish)
Same four-field format.

## Not-observable
List what you could not verify and what would be needed to check it
(e.g. "brand color hex values not embedded in .docx file; export as
PDF or provide a color-checked screenshot").

## Guidelines gaps
One section listing any issue in the document that the guidelines
do NOT cover clearly. These are prompts for us to improve the
guidelines file.

Rules for the review:
- quote exact phrases, do not paraphrase
- never invent a rule that is not in the guidelines file
- if a rule contradicts itself, flag it in the "Guidelines gaps"
  section rather than guessing
- prefer fewer, sharper fixes over an exhaustive list

Iterate on the prompt and on the guidelines file together. Every “Guidelines gaps” item is a tighten-the-guidelines task, not a reviewer failure.

Step 3: Wire It Into How Work Actually Flows

The reviewer is useless if it only runs when someone remembers. Pick one of these patterns and stick with it.

Pattern A: Review-On-Mention In Telegram Or Slack

From the team chat where docs are shared for review:

🎯 Try it yourself

When someone shares a document in this chat and asks for a brand
review, run the brand reviewer against our saved brand-guidelines.md
and post the review back in the same thread. If the document is a
URL, fetch and review the page body. If the verdict is "do-not-ship",
prepend the response with a single red flag in the first line.

Pattern B: Scheduled Sweep Of A Shared Folder

If your team drops drafts into a shared folder:

🎯 Try it yourself

/schedule add "0 9 * * 1-5" Every weekday morning, review every document added to the "for-brand-review" folder since yesterday. Use our saved brand-guidelines.md. For each document, return the verdict, the must-fix list, and a one-sentence summary. Skip documents already reviewed.

Pattern C: One-Shot Pre-Send Check

For the founder or a single reviewer, the simplest pattern is a manual kickoff right before hitting send. No schedule, no chatbot, just a habit:

🎯 Try it yourself

Review the document I just uploaded against our saved
brand-guidelines.md. Use the standard five-dimension review with
verdict, must-fix, should-fix, nice-to-have, not-observable, and
guidelines-gaps sections. Be strict — I would rather fix one thing
twice than miss it once.

The Rules That Matter

Three rules keep the reviewer honest.

The reviewer only knows what is in the guidelines file. If the reviewer misses an obvious issue, the fix is almost always to add a line to brand-guidelines.md, not to make the prompt longer. A sharp one-page guideline beats a vague ten-page one every time.

Quote, don’t paraphrase. The review is only useful if you can jump to the exact sentence and change it. Insist on verbatim quotes; reject summaries.

Separate must-fix from polish. If everything is important, nothing ships. The ordered sections exist so a tired reviewer can fix three things in ten minutes and move on.

Useful Variations

• Tone-only pass for a social post. Strip the checklist to voice and terminology. A LinkedIn post does not need a compliance section.
• Localized review when you publish in multiple languages. Add a “Translation quality” dimension and ask for flagged literal translations that read awkwardly to a native speaker.
• Legal-lite pass for investor material. Add a sixth dimension covering forward-looking statements and any performance claims that need a safe-harbor footer.
• Diff-mode review. Ask for a review of only what changed between version N-1 and N. Useful in late-stage doc editing where a full review is noise.
• Brand-fit scoring for external content. Point the reviewer at a press article, partner co-marketing draft, or analyst report to check whether external coverage matches how you actually talk about yourself.

Production Tips

• Keep brand-guidelines.md under 500 lines. Long guidelines get ignored; short ones get enforced.
• Every time the reviewer flags a “Guidelines gap”, spend two minutes updating the file. After a month the gaps dry up.
• Do not run the reviewer against its own output. Reviews are not copy-edited documents; they are worksheets.
• Review the reviewer monthly. Scan ten reviews and ask: did it miss anything real? Did it flag anything that was actually fine? Both are prompts to tighten either the guidelines or the prompt.
• When a “must-fix” slips through to a published document, treat it as a guideline bug, not a reviewer bug. Write the rule that would have caught it next time.

Going Further

• Office Dependencies
• Commands
• Adaptive Skills