Claude, Applying Brand Guidelines
A look at the applying-brand-guidelines Anthropic Skill, which reads a written spec instead of exercising taste, runs the same three steps on any document, and checks its own output with validate_brand.py every time.
The applying-brand-guidelines Skill answers a specific question: when Claude brands a document, is it exercising some general design sense, or following a written spec? The video makes the case for the latter by showing what the Skill's folder actually contains and how it behaves when run twice on the same input.
A skill is a folder, not a trained instinct
applying-brand-guidelines is not a vague brand feel baked into the model. It is a folder Claude reads before it works, containing apply_brand.py, validate_brand.py, a REFERENCE.md, and a four-kilobyte SKILL.md that holds the full instruction set in plain language, with no hidden logic sitting outside it. Claude reads the file, then acts, the file is the program, and the instructions live in a numbered step section that gets executed in order, with no branching unless a step explicitly calls for it.
What the anchor example shows
The video's anchor example hands the Skill a slide deck. It reads the deck, applies the specified colors and fonts, and returns a branded deck, the same three steps every run. To demonstrate that this is a fixed procedure rather than a judgment call, the same deck is handed to the Skill twice, and apply_brand.py returns identical branding both times. That repeatability is presented as the direct payoff of the file being the program: no drift, no different interpretation on a second pass.
Where the scope limit shows up
The more interesting case is the reverse. When a document falls outside the Skill's stated scope, the example given is external communications only, the Skill still runs the same steps against material SKILL.md never specified for. It does not detect that the document is out of scope and refuse; it applies the same procedure regardless. The check that catches mistakes is validate_brand.py, which flags wrong colors, wrong fonts, or scope problems the same way every time, after the fact rather than by pre-filtering input.
What this establishes about the Skill's behavior
Put together, these two behaviors describe a system that is consistent rather than adaptive: identical input produces identical output, and the Skill does not exercise discretion about whether a document is a good fit for it. Applying brand guidelines this way is not Claude developing its own sense of style, it is SKILL.md acting as a spec that Claude reads, applies, and validates the same way against whatever document it is handed.
What this video does not claim
The video does not render a verdict on whether the Skill's SKILL.md should cover more document types or brand elements than it does, that is a separate question from describing the mechanism. It also does not name specific color codes, font names, or layout rules beyond what the source material already describes, since the underlying SKILL.md was not available to expand on. And it does not claim the Skill decides what counts as "on brand", only that it runs the same steps and the same check regardless of whether a document fits its stated scope.
Key takeaways
- The Skill's behavior comes entirely from a readable
SKILL.md, plusapply_brand.pyandvalidate_brand.py, no hidden logic. - Instructions are executed as a linear sequence of steps, without branching unless a step calls for it.
- The same input, run twice, produces identical branded output.
- Documents outside the Skill's stated scope still get the same steps run against them, the Skill does not self-limit by scope.
validate_brand.pyis the check that catches wrong colors, wrong fonts, and scope mismatches after the fact.
Who this is for
Anyone deciding whether to trust an AI branding tool with team documents, and anyone who wants to see how an Anthropic Skill's folder structure, SKILL.md plus scripts, produces consistent, checkable behavior instead of case-by-case judgment.
Chapters
Full transcript(auto-generated, with timestamps)
The naive framing: "match our brand"
[0:00]Someone wonders how you'd get Claude to match your brand, but matching a vibe isn't it. This skill applies a written spec. So, how does that work? A skill is
A skill is a folder Claude reads — not trained on it
[0:08]A folder Claude reads before it works, not something it's trained on. This one is applying brand guidelines. Apply brand pi, validate brand pi, reference MD in a 4 kilobyte skill. MD holding the full instruction set plain language, no hidden logic. Claude reads the file then acts. The file is the program. The
The anchor: three steps, a slide deck through apply_brand.py
[0:26]Instructions sit in a step section. Read skill MD executing each step in order to return the result linear, no branching unless a step says so. Watch the anchor. Hand the skill a slide deck. It reads the deck, applies your colors and fonts, and hands back a branded deck. Same three steps every time. Same input, same
The anchor returns: same input twice, then the scope limit
[0:43]Output every run. Hand it the identical deck twice and apply brand pi returns the identical branding both times. That's the payoff of a file being the program. But, the reverse holds too. Hand it a document outside the stated scope, external communications only, and it still runs the same steps against material the skill MD never specified. Validate brand pi is the check. Wrong colors, wrong fonts, wrong scope, all caught the same way. Applying brand
Carry-out
[1:06]Guidelines isn't Claude developing its own sense of style. It's skill. MD, a spec Claude reads, applies, and validates the same way against whatever document you hand it. Your turn. Here's
Your turn
[1:17]The prompt. Read it with me. I want to apply brand guidelines to my team's documents. Read the applying brand guidelines skill and walk me through what you will do before you do it. Watch for that walk-through explaining first is what surfaces which step skill MD actually specifies and where the scope stops. Liam in for bear. Claude applying brand guidelines. Liam
Outro
[1:37]In for bear.
