Quick Answer
To turn an SOP into an AI skill, keep the SOP in your knowledge base as the detailed how, and write a short skill that points to it. The skill names the job, the inputs, the steps at a glance, the output and the checks, plus the decisions the SOP leaves to experience. Your company context supplies the tone and terms. Then test it with three cases.
Why an SOP Isn't a Skill Yet
An SOP is written for people. It assumes the reader knows where the templates are, which client is which and when to ask a colleague. A new starter fills those gaps by asking. An AI fills them by guessing.
A skill is written for the AI. Think of the SOP as the manual and the skill as the job card that says which pages of the manual to follow. Microsoft draws a similar line in Copilot Studio, where knowledge is material an agent can reference and skills are task-specific instructions (Microsoft: Skills overview for Copilot Studio agents).
Each part has its own job:
| SOP | Skill | Company context | |
|---|---|---|---|
| Written for | People | Your AI | Your AI, under every skill |
| Holds | The full procedure, with detail and exceptions | When to use it, inputs, steps at a glance, output and checks | Who you are, your terms, tone and formats |
| Lives in | Your knowledge base | Wherever your AI's skills are kept | Your company and team rules |
| Changes when | The process changes | The job or its output changes | The business changes |
Keeping them apart gives you one source of truth. The SOP's owner can update a step without rewriting the skill, and the skill stays short. The open standard suggests keeping SKILL.md under 500 lines and moving detail into files the AI opens only when needed (Agent Skills specification).
How to Turn an SOP Into a Skill
Start with an SOP that's current and used. If it's out of date, fix it first, or the skill will spread the old process to every AI your team uses.
Name the job and the trigger. Write one sentence with the task and its result, then list the words people use when they ask for it. That becomes the skill's description.
Pull out the inputs. List what the AI needs before it starts, and mark the ones it must never guess, such as fees, dates and contact names.
Summarise the steps and point to the SOP. List each step in a line and name the SOP section that holds the detail. Don't copy the SOP's wording in, or the two will drift apart.
Write down the decisions the SOP leaves to experience. "Use your judgement" isn't an instruction an AI can follow. Say what counts as a complex client, and when a missing document stops the job.
Define the output. Name each document or message the job produces, its sections and who reads it.
Add the checks. End with a short list the AI runs before finishing. If the job sends anything, add an approval step here too.
Leave tone and terms to company context. The skill shouldn't repeat who you are; your company context covers that.
Worked Example: The Client Onboarding SOP
Harbourview Accounting, a fictional accounting practice, has a six-page Client Onboarding SOP in SharePoint, covering the engagement letter, identity checks, the information request, the client file and the welcome email. New staff follow it well. Asked to "onboard this client", an AI wrote a friendly welcome email and skipped the identity checks.
The SOP stays where it is. The Client Onboarding Skill, client-onboarding, fits on a page:
- Description: Prepares the onboarding pack for a new client using the Client Onboarding SOP. Use when someone asks to onboard a client, set up a new client or send an engagement letter.
- Inputs: the client's legal name, entity type, main contact, agreed services and fee basis. Never guess the fee or the services.
- Steps: follow the SOP in order: engagement letter (section 2), identity checks (section 3), information request (section 4), client file (section 5) and welcome email (section 6).
- Decisions: for a trust, use the trust checklist in section 3. If identity documents are missing, stop and list what's needed.
- Output: a draft engagement letter from the current template, an information request list, a draft welcome email and a note of anything outstanding.
- Checks: the fee matches the agreed basis, every service in the letter was agreed, and nothing goes to the client until a person has reviewed it.
Company context covers the rest: Harbourview's name, New Zealand spelling, how it signs off emails and that it says "clients", never "customers". When the practice updates its engagement template, the template changes and the skill doesn't.
The first test showed the AI drafting a welcome email for a client with missing identity documents. The fix was one line in the skill: "Don't draft the welcome email until the section 3 checks are complete." The second test passed.
Try It: Turn One SOP Into a Skill
- Choose an SOP your team follows at least weekly, and check it's current.
- Write the one-sentence job and result, and list the words people use to ask for it.
- List the inputs and mark the ones the AI must never guess.
- Summarise the steps in a line each, naming the SOP section for the detail.
- Write the decisions the SOP leaves to experience.
- Name the outputs and add three to five checks.
- Confirm your AI can open the SOP and templates where they live.
- Run the three tests below.
Test It: Three Cases
- A normal request. Onboard a fictional client with complete details. Compare each step against the SOP, not just how polished the drafts look.
- Incomplete input. Leave out the fee basis. The AI should ask for it or mark it as missing, not pick a figure.
- An unrelated request. Ask for a reminder to an existing client about an overdue return. The onboarding skill shouldn't take over.
Then change one detail in the SOP, such as a document on the identity checklist, and run the normal case again. The AI should follow the updated SOP without anyone editing the skill.
Where Contexta Fits
A skill tells your AI what to do. Your knowledge base shows it how, in detail. Your company context tells it who it's for. The Contexta harness brings them together, in every AI.
The Client Onboarding SOP stays in your knowledge base, in your own SharePoint or Google Drive; Contexta doesn't store or index it, and your AI reads it there with its own access. The skill holds the job card and points to the SOP. Company context covers Harbourview's terms, tone and formats, set for the company and adjusted for each team. The client services space routes the work to its own folders, with its own rules and audience.
Contexta's built-in knowledge base skills, built on best practice, help your AI turn documents such as meeting notes and process write-ups into SOPs, decisions and lessons, and find them when a job needs them. When you publish the skill, connected AIs fetch it on next use. Version history shows who changed what and when, with compare and restore. Contexta is tested with Claude, ChatGPT and Cursor; see How It Works.