Beyond individual agents, your organisation's **Library** holds three reusable building blocks that shape how agents behave: **skills**, **rules**, and **memories**. They differ in who controls them and how strict they are — guidance an agent can use, instructions it must always follow, and know-how it remembers. Each can be scoped to all agents or specific ones, and to your whole organisation or specific projects.

## Skills

A skill is *reusable AI guidance applied to your organisation's agents* — markdown guidance layered onto an agent's system prompt. Use one to package a way of doing something you want available across agents, rather than re-explaining it each time.

Skills come from two places: the ones your organisation authors, and a catalogue of **system skills** Regnora maintains. You can **fork** a system skill to create an editable organisation copy. When configuring an agent you choose which skills it should use, and you can scope a skill by agent type and by project.

**Manage:** [Skills](https://app.regnora.com/library/skills)

## Rules

A rule is an *always-on instruction the agent follows every run* — authored by your team, and something **the agent cannot change**. Rules are injected into the system prompt on every matching run, which makes them the place for non-negotiables: a tone of voice, a compliance constraint, a phrasing your organisation always uses.

Rules are organised by a path (like `coding/tone-of-voice`) and scoped by agent type and project. Disabling a rule keeps it in the bank but excludes it from every agent run, so you can retire one without losing it.

**Manage:** [Rules](https://app.regnora.com/library/rules)

## Memories

A memory is *persistent know-how the agent remembers between conversations*. Unlike rules, memories are collaborative: an agent **proposes** changes from what it learns during its runs, and your team **reviews and accepts** them into the shared bank. You can also add a memory directly.

Proposed changes — to create, update, or delete a memory — wait under **Pending review**, where an admin can accept, reject, or edit them before they take effect, and a history lets you revert to an earlier version. Disabling a memory means the agent will no longer recall it. Like the others, memories are scoped by agent type and project.

**Manage:** [Memories](https://app.regnora.com/library/memories)

## How they attach to agents

The three differ in how they reach an agent. **Skills** are chosen per agent — you pick the ones an agent should use. **Rules** and **memories** apply automatically to every agent that matches their scope, with no per-agent wiring: set the scope and they're in effect for the agents it covers.