---
title: "Instructions, Skills and Memory"
url: "https://plugpress.io/docs/guidance-and-skills.md"
canonical: "https://plugpress.io/docs/guidance-and-skills/"
published: "2026-08-01"
modified: "2026-08-26"
author: "Fahim"
description: "Teach your AI how you work. Standing instructions apply to every session, and installable Skills give it a step-by-step playbook for your tasks."
tags:
  - "Saddle"
---

# Instructions, Skills and Memory

**Permissions decide what your AI is allowed to do. These three decide how it does it.** The **Saddle → Instructions** screen holds standing instructions that go to every session and **Skills** — installable playbooks that teach an assistant a whole workflow. **Saddle → Memory** holds the facts that survive between sessions.

![Saddle Instructions screen showing the owner instructions box and two installed Skills with on and off switches](https://plugpress.io/wp-content/uploads/2026/08/saddle-guidance-skills.png)

## What are standing instructions?

A plain-language note that every connected assistant reads at the start of every session. Use it for the rules you’d otherwise repeat in every conversation:

- “Always save new posts as drafts for me to review.”
- “Our brand voice is direct and unfussy. No exclamation marks.”
- “Never touch anything in the Legal category.”
- “Use British spelling.”

Write it, click **Save instructions**, done. It applies to every app you’ve connected, not one at a time.

Instructions are guidance, not enforcement. If a rule really matters — “never delete anything” — set it in [Permissions](/docs/permissions-access-levels/) as well, where it’s enforced rather than requested.

## What is a Skill?

A Skill is a Markdown file describing a workflow, installed on your site, that connected assistants discover and follow automatically when a task matches it.

Where an instruction is a rule, a Skill is a procedure. A publishing Skill might specify: research the search intent first, use question-phrased headings, answer in the first paragraph, add internal links, generate the SEO meta, run the readiness check, and only then publish. When you ask for an article, the assistant follows that — instead of whatever it would have done by default.

## How do I install a Skill?

1. Write or download a `.md` file with a frontmatter block at the top.
2. On **Saddle → Instructions**, click **Add skill** and choose the file.
3. It appears in the list with a switch, a **View** link, and **Delete**.

The frontmatter needs a name and a description; `when_to_use` is optional but helps the assistant judge when the Skill applies:

```
---
name: write-blog-post
description: Our house process for writing and publishing a blog post.
when_to_use: When asked to write, draft, or publish an article.
---

1. Check what we've already published on the topic.
2. Draft with question-phrased H2s and an answer-first opening.
3. Link to at least two related posts.
4. Save as a draft and tell me the URL — never publish directly.
```

Installing a Skill with an existing name replaces that one, so updating is just re-uploading.

## Why can only I install Skills?

Because a Skill is text an AI will follow. If an assistant could write its own Skills, it could write its own instructions — and anything that can quietly rewrite its own rules is a problem, not a feature.

So the split is deliberate: **agents read Skills, owners install them.** There is no tool that creates or edits one. Uploading requires a WordPress administrator.

A Skill also never grants capability. It can only orchestrate tools that are already switched on at your access level. A Skill that says “delete the old drafts” still hits the [approval gate](/docs/safe-deletions/), and still fails outright if the delete tools are off.

## How does the AI know which Skill to use?

Every session receives a short index — each Skill’s name, description, and when-to-use line. That’s small enough to send every time. When a task matches one, the assistant fetches the full text on demand.

Practically: keep the description specific. “Our house process for writing and publishing a blog post” gets matched. “Content stuff” doesn’t.

## What is Memory?

Instructions and Skills are things you write once and leave in place. **Memory** is for facts about your site that accumulate — the things you’d otherwise re-explain at the start of every conversation.

It lives on its own screen, **Saddle → Memory**, and has two halves that behave differently:

- **Notes you write yourself** ride along with every session, like a shorter, more factual version of your standing instructions. “Our product pages all use the Feature Grid pattern.” “The client signs off before anything in /case-studies/ is published.”
- **Notes the assistant wrote** stay *recall-only* until you pin them. An assistant can record what it learned and look it up later, but nothing it wrote about your site is pushed into every future conversation unless you decide it earns that place.

That asymmetry is the point. Memory an agent can silently promote into every session is a way for a wrong fact to become permanent; this way, the things that ride every conversation are the things you chose.

## Limits

| Limit | Value |
| --- | --- |
| Skills per site | 50 |
| Length of one Skill | 30,000 characters |
| File format | Markdown (`.md`) with frontmatter |
| Who can install | Administrators only |

Skill bodies are stored exactly as you wrote them — installing only removes invisible control characters, nothing else. Placeholders in angle brackets, like `<id>` or `<page-url>`, are part of your instructions and reach the assistant verbatim, so use them freely.

## FAQ

### What’s the difference between instructions and Skills?

Instructions are short, always-on rules that apply to everything. Skills are longer procedures loaded only when relevant. Rule of thumb: if it applies to every task, it’s an instruction; if it’s “how we do X”, it’s a Skill.

### And what’s the difference between an instruction and a memory?

An instruction is a rule you’re setting. A memory is a fact about your site. “Always save posts as drafts” is an instruction; “our pricing page is built with the Feature Grid pattern” is a memory.

### Can I turn a Skill off without deleting it?

Yes — each has its own switch. Off means it isn’t offered to assistants at all.

### Do Skills work across all my connected apps?

Yes. They’re stored on the site, so every connected assistant sees the same set.

### Can plugins ship their own Skills?

Yes — Saddle Pro includes a Divi page-building playbook, for instance. A Skill you install with the same name takes precedence, so you can always override one.

### Can a Skill make the AI do something I’ve blocked?

No. Skills describe process; Permissions decide power. A Skill asking for a blocked tool simply fails.

## Related guides

- [Permissions and access levels](/docs/permissions-access-levels/)
- [Editing content and design with AI](/docs/editing-content-and-design/)
- [Safe deletions and the approval gate](/docs/safe-deletions/)
- [Getting started with Saddle](/docs/getting-started/)