The Scenario.
You have a 47-page standard operating procedures document sitting in a shared drive somewhere. Last updated eight months ago. Half your team hasn’t read it. The other half read it once and forgot it.
Now imagine that same process knowledge living inside an AI agent that every employee on your team uses daily. Not as a document they have to go find and read. As executable instructions the agent follows automatically, every time, producing consistent outputs across your entire organization.
That’s the shift happening right now in businesses that are serious about AI. They’re moving from SOPs to skills. And the companies that figure out how to govern that transition are going to operate at a completely different level than the ones that don’t.
We Made the Switch. Here’s What It Looks Like.
At Cirrius Solutions, we’ve been rolling out what we call our Cirrius Ops Agent. It’s a locally based employee productivity agent built on Claude Code that runs on every team member’s machine. Not a chatbot. Not a copilot sitting in a sidebar. A full agent that understands our business processes, connects to our tools, and executes real work.
We currently have 27 active skills deployed across the company. That number grows weekly. These skills cover everything from how we organize project notes and run weekly reviews to how we interact with ClickUp, draft client communications, and build business value assessments for enterprise engagements.
But here’s the thing we learned fast: skills without governance become chaos. If every employee is building and deploying their own skills with no structure, you just traded one mess (dusty SOPs nobody follows) for another (a wild west of inconsistent agent behaviors across your team).
What Skills Actually Are (And Why They Matter More Than Prompts)
A skill is a set of structured instructions, combined with supporting context and reference materials, that enables an AI agent to accomplish a specific task. Think of it as a repeatable, executable process definition that’s readable by both humans and agents.
That last part is critical. A well-built skill replaces both your SOP documentation and your center of excellence for that process. One artifact to maintain instead of two. Your employees can ask the agent “how do we do business value assessments here?” and get an answer drawn from the same skill that the agent uses to actually build one.
Skills aren’t just a Claude Code concept, either. They’re quickly becoming the common language across agent platforms. Whether you’re running Claude Code, Codex, or building on an agent framework, skills (or something architecturally similar) are how you define what your agent can do. If you’re investing in AI agents for your business, skills are the building blocks you need to get right.
And skills go well beyond development. We’re seeing massive productivity unlocks at the reporting level, task management level, and operational level. This is not a developer-only tool. This is a business operations tool.
The Three-Tiered Governance Structure
After some early lessons in what “the wild west” actually looks like when you let skills grow unchecked, we built a three-tiered governance structure. It balances centralized control (so the business operates consistently) with individual freedom (so employees can tailor their agent to how they actually work).
Tier 1: Individual Skills. Any employee can build skills for their own workflow. Experimenting, optimizing, figuring out what works for their specific role. This is where innovation happens. When an employee builds something that works, they push it to a shared repo for review.
The QA Gate. Before any individual skill can be shared beyond that one person, it goes through a security and quality review. A dedicated team evaluates the skill for security risks (does it expose anything it shouldn’t?), process alignment (does it follow how we actually do things?), and quality of output. If it needs changes, the employee gets specific feedback. This gate prevents one person’s experiment from becoming the whole company’s problem.
Tier 2: Department or Role Skills. These are skill packs for specific functions. Sales has a skill pack. Managed services has a different one. Project delivery has another. This middle tier exists because not every skill needs to be company-wide, but some skills are too valuable to stay with one person. A sales team skill for proposal generation doesn’t need to go to the engineering team. But every account director should have it.
Tier 3: Company-Wide Skills. These are the baseline. Every employee gets them. Note-taking conventions, weekly reviews, task management, communication standards. These define the operational floor for the entire organization. They ensure that regardless of department or role, everyone’s agent operates with the same foundational behaviors.
The key principle: employees should have exactly the skills they need, plus the freedom to individualize, without bloat. The more irrelevant skills an agent has loaded, the higher the chance of inaccurate tool calls or confused behavior. Lean and relevant beats comprehensive and noisy every time.
Inside the BVA Builder: What a Real Skill Looks Like
Theory is easy. Let’s look at an actual production skill to see what this means in practice.
Our Business Value Assessment (BVA) Builder is one of the most impactful skills in our deployment. As a Salesforce consulting firm, proving the business value of the solutions we implement is core to what we do. We build BVAs during the sales process, maintain them throughout engagements, and assess them after go-live. They directly impact whether clients continue working with us.
Before this skill existed, building a BVA was a manual, multi-hour process requiring significant experience and industry knowledge. Now, an account director fires the skill and the agent walks through a structured flow:
- It asks for inputs: existing BVA spreadsheet (if one exists), the client’s statement of work, client name, vertical industry, and any relevant meeting transcripts or pre-sales notes.
- It generates a multi-tabbed spreadsheet mapping executive priorities to strategic goals, operational objectives, capabilities, and measurable metrics.
- It deposits the output into Google Drive and can create formula fields in Salesforce for ongoing tracking.
- A human reviews, refines, and approves before anything goes to a client.
But the skill itself isn’t just a single markdown file with instructions. It’s a folder with five critical reference documents that make it actually work:
- A gold-standard example of what a great BVA looks like, formatted the way we want it done.
- A schema and lexicon defining key terms, fields, and how we talk about business value internally.
- An industry checklist with common value patterns for the verticals we serve, so the agent draws on real domain knowledge instead of generic suggestions.
- Quality checks that must pass before the output is ready for human review.
- Salesforce formula patterns from past engagements, so metric tracking configurations are grounded in what actually works.
This is human-agent collaboration at its best. The agent handles the assembly, the formatting, the cross-referencing of industry patterns, and the initial quality checks. The human brings the relationship context, the judgment calls, and the final review. Neither could do this as well alone. Together, a process that used to take hours produces a strong first draft in minutes.
We’re releasing a genericized version of this skill on our GitHub so you can see the structure and adapt it for your own business.
Why This Is a “One-Time Pain, Many-Time Gain” Investment
Building skills takes real effort. The BVA Builder required us to formalize three years of process knowledge into structured, agent-readable formats. We had to go deeper on documentation than we ever had before.
But that’s exactly the point. You do this work once, and every employee benefits from it permanently. Every new hire gets the accumulated process knowledge of your best performers on day one. Every engagement follows the same proven approach. Every output meets the same quality bar.
The alternative is worse. If you don’t build this structure, your employees are still going to use AI agents. They’re going to build their own workflows, their own prompts, their own approaches. You’ll just have no visibility into what they’re doing, no consistency in outputs, and no way to improve the process centrally. You’ll spend that time anyway. You’ll just spend it cleaning up instead of building up.
Where to Start
- Identify your top 5 repeatable processes. What do multiple people in your organization do repeatedly that follows a predictable pattern? Start there. Not with the hardest thing. With the most common thing.
- Pick one and build the skill. Document the process as structured instructions with clear inputs, steps, and expected outputs. Include reference examples of what “good” looks like. Test it with the team that actually does the work.
- Stand up a simple governance structure. Even if you start with just two tiers (individual and company-wide), establish the principle early: skills get reviewed before they spread. A shared repo with a review step is enough to start.
- Measure the output, not the activity. Skills should be judged by the quality and consistency of what they produce, not by how clever the instructions are. If the output isn’t better than what your team was producing manually, the skill needs more work.
We Build This
Cirrius Solutions doesn’t just talk about AI agents. We run them across our own company every day. Our Cirrius Ops Agent, our skills governance framework, and our production skills like the BVA Builder are the same patterns and approaches we bring to our clients.
This post is based on a recent episode of our podcast where Tim Harting, Gavin Franklin, and Greg Banks go deeper on skills governance, the BVA Builder, and what we’ve learned rolling out the Cirrius Ops Agent. Watch the full episode here.
If you’re ready to move from SOPs to skills, from individual experimentation to structured deployment, from AI curiosity to AI operations, let’s talk about what that looks like for your business.
