The Skills System That Keeps Every Line of AAA Content On-Brand

Two Skills, Not One Editor

Run a generic prompt through a general model and it writes fine, forgettable copy. Every piece of AAA content runs through two Skills before anyone reads it. The first sets voice. The second checks mechanics.

A Claude Skill is a packaged set of instructions a model loads before it writes. Ours are checked into this site’s own repository. Any session that touches AAA content loads them automatically, including the one that wrote this article.

What Each Skill Actually Enforces

The first Skill, aaa-brand-guidelines, sets the voice. It tells the model to lead every sentence with a verb and name products correctly: Active GEO, not “our GEO tool.” It also trades vague claims for a real number or a named system. No exclamation points. No “revolutionary.” No “game-changing.”

The second Skill, aaa-writing-guide, runs after the first as a mechanics pass. It caps sentences at roughly 10 to 20 words, and bans semicolons and em dashes outright. It also flags a list of over 60 banned words and phrases on sight, including “leverage,” “ensure,” “seamless integration,” and “paradigm shift.”

Between them, the two Skills catch what a human editor catches on a second read, before a draft ever reaches one.

A Real Before and After

Here is one example, straight from the writing Skill’s own file, unedited.

Five banned words sit in that one sentence, on top of a stacked clause structure the guide flags on sight.

Same claim. Half the words. One real number instead of an adjective doing the work a number should do.

Why This Matters More Than Voice

Brand voice is one use for a Skill. It is not the only one. The same system can codify anything AAA needs repeated correctly, not just how a sentence reads.

• Onboarding. A new session, human or AI, loads the Skill and starts on-brand immediately. No ramp-up call required.
• QA that survives a rewrite. The visual PDF-verification technique used across every Field Notes export is a candidate for its own Skill, so the check runs the same way regardless of who runs it.
• Fewer repeated mistakes. A past session that pushed straight to production from the wrong branch became a documented gotcha, not just a memory that fades by the next one.
• Consistency at volume. Ten writers and one AI session can produce the same voice. The rules travel with the prompt instead of living in one person’s head.

Where Else This Pattern Applies

Content is one use of a Skill. Most repeatable judgment calls work the same way. None of the four examples below are real AAA client work. They show the shape a Skill can take in a function that is not writing.

• Sales. A Skill that puts one pricing-disclosure line on every proposal, regardless of which rep drafted it.
• Support. A Skill that flags any ticket mentioning “refund” or “legal” for a human to review before an AI drafts a reply.
• Finance. A Skill that fixes one column order and one rounding rule across every board deck, so a number does not change shape depending on who built the slide.
• Engineering. A Skill that encodes a postmortem template, so every incident review answers the same five questions in the same order.

Each example above is illustrative, not a client engagement. The AAA content Skill described earlier is the one instance of this pattern that is real and running today.

What This Article Proves

This piece ran through both Skills before it shipped. That is checkable, not a claim. Anyone on the AAA team can open the same two files in this repository and see the same rules applied here.

Active Forge is the name for this practice area. This piece is the first build under it. The next is a real agent: one that plans, calls tools, and shows its own trace, rather than a chatbot with extra steps.