# Give agents a playbook they can use

Keep brand rules and working instructions in reusable skills, with clear versions, owners and review steps.

By Product Team · Published June 1, 2026 · Updated October 8, 2026 · AI & marketing

Can you imagine logging into a cutting-edge generative AI platform to launch a critical product campaign, only to realize you have to spend the first twenty minutes teaching the AI who you are?

With the rapid, incredibly recent rise of Large Language Models (LLMs) and generative AI in the enterprise workspace, companies worldwide have run headfirst into a frustrating bottleneck: the "blank slate" problem. We are suddenly surrounded by highly intelligent, capable AI agents, yet every single interaction feels like hiring a brilliant consultant who has absolutely no idea how your specific company operates. You find yourself repeatedly copying and pasting the same brand guidelines, compliance rules, and target audience definitions into every single prompt box.

Reusable skills let teams keep instructions and references together. They still need owners, version control and testing as the business changes.

While our agents have always utilized native, behind-the-scenes skills to perform their tasks, we are now handing you the keys to the engine. You can now build, upload, and run private, highly customized skills for each of our four autonomous agents. It is time to turn general-purpose LLMs into your own dedicated department of specialized enterprise experts.

## The Mechanics of SKILL.md: Deep Context Without the Bloat

Before we look at our specialist agents, let's explain exactly what a skill actually is.

At its core, a skill is a lightweight directory containing a single plain-text file called SKILL.md written in clean, human-readable Markdown, packaged alongside any supporting datasets, reference documents, or script files.

If you feed an entire enterprise library to an AI model at the start of every chat, the model gets overwhelmed. It suffers from "context bloat" — slowing down, driving up API costs, and occasionally losing track of the instructions.

To solve this, Random Truffle uses a highly sophisticated architecture called Progressive Disclosure.

The agent only ever pays the cognitive cost for what it actually needs in the moment. Your full enterprise library can live in the vault; only the relevant blueprint comes off the shelf when a real task triggers it.

## A Specialist Skill for Each of Your Four Agents

Random Truffle's autonomous marketing team is built around four agents — Aura, Nexus, Radian, and Prism. Each one runs an entirely different part of your marketing operation, and each one now accepts its own portfolio of custom skills.

### Aura — Brand Voice & Creative Guardrails

Aura is the creative agent. Give her a skill called brand-voice.md containing your tone-of-voice rules, banned words, claim restrictions, and product naming conventions — and every headline, body line, and email subject she produces will already be on-brand and legally clean. No more 14-step copy reviews. The compliance team becomes a coach, not a bottleneck.

### Nexus — Your Data Model in Plain English

Nexus is the data agent. Upload a skill that documents your semantic layer: how revenue is defined in your warehouse, which BigQuery tables hold pLTV, what "active customer" means in your business, and the SQL conventions your team enforces. Nexus stops guessing and starts answering questions the same way your senior analyst would — citing the same metrics, using the same joins.

### Radian — Channel Playbooks That Travel With You

Radian runs activation across Google, Meta, TikTok, and the long tail of programmatic. Hand her a skill with your bidding philosophy, geo strategy, exclusion lists, and creative rotation rules per channel. Suddenly, every campaign she launches follows the same playbook your best media buyer wrote — at machine speed, across every market, without drift.

### Prism — Measurement With Your Priors Baked In

Prism is the measurement agent that owns your MMM and incrementality work. A custom skill can encode your Bayesian priors, your saturation curves by channel, the lift tests you've already run, and the business constraints the model must respect. Prism stops giving you generic econometrics and starts giving you measurement that already agrees with what you know to be true about your business.

## From Generic LLM to Enterprise Specialist

The promise of generative AI was never "a chatbot that knows everything about everything." It was a workforce that compounds your institutional knowledge instead of starting from zero every Monday morning.

Custom skills are how you get there. Write your playbook once. Version it like code. Let four autonomous agents apply it across thousands of decisions a week — every one of them grounded in your strategy, your data, and your guardrails.

### "A skill turns a brilliant generalist into your specialist — without the onboarding."

Custom SKILL.md support is rolling out to all Random Truffle workspaces this week. Bring your playbook. We'll bring the agents.

---

Source: https://randomtruffle.com/research/make-ai-yours-how-custom-skills-unleash-random-truffles-agents