# Build Agent Skills That Load on Demand

> Save repeatable agent commands as Skills so they load on demand and persist across reloads.

- Source: https://www.taskade.com/learn/agents/agent-skills
- Section: AI Agents
- Updated: 2026-08-26
- Index of all pages: https://www.taskade.com/learn/llms.txt

## Overview

Skills are saved, named commands you can attach to any AI agent. Type a slash, pick the Skill, and the agent runs the playbook you defined. Skills load on demand, so an agent's prompt stays lean even when you have dozens of them ready.

> **TL;DR:** Save repeatable agent instructions as Skills. Call a Skill with a slash command whenever the same task returns. Skills keep prompts focused and travel with shared agents. Pair them with the [Bash Sandbox](https://www.taskade.com/learn/agents/bash-sandbox) when a task also needs approved scripts.

Think of a Skill as a card the agent keeps in a drawer. It stays out of the way until the task matches, then the agent pulls it out, runs it, and puts it back.

```text
┌────────────────────────────────────────────────────────────────┐
│  Skill  ›  /qa-checklist                                       │
├────────────────────────────────────────────────────────────────┤
│  Name         /qa-checklist                                    │
│  Loads when   the task mentions "review" or "QA"               │
│  What it does runs your 8-step quality pass on a draft         │
│  Status       parked  ·  loads on demand  ·  no prompt cost    │
├────────────────────────────────────────────────────────────────┤
│  Agent context (right now)                                     │
│  [ /qa-checklist  parked ]  [ /draft-followup  parked ]        │
│  [ /summarize-thread  ACTIVE: loaded for this task ]           │
└────────────────────────────────────────────────────────────────┘
```

Here is the full loop, from a new task to the Skill unloading again.

```mermaid
flowchart LR
    A[Agent gets a task] --> B[Matches a Skill by name or trigger]
    B --> C[Loads the Skill on demand]
    C --> D[Uses the Skill to do the work]
    D --> E[Unloads, context stays lean]
    style A fill:#0d1b2a,color:#7dd3fc,stroke:#38bdf8,stroke-width:1.5px
    style B fill:#1a0d24,color:#ce93d8,stroke:#6a1b9a,stroke-width:1.5px
    style C fill:#2a1225,color:#ff8fa3,stroke:#ff2d60,stroke-width:2px
    style D fill:#0d1a0e,color:#86efac,stroke:#22c55e,stroke-width:1.5px
    style E fill:#1a140a,color:#ffb74d,stroke:#e65100,stroke-width:1.5px
```

## When to use Skills

A Skill is a great fit when:

- You repeat the same prompt or sequence multiple times a week
- You want an agent's playbook to stay short while keeping many capabilities
- You want teammates to invoke the same workflow with one slash-command
- You want to swap a Skill in or out without retraining the agent

For example, a construction project manager saves a `/daily-report` Skill that pulls today's logged hours, open punch-list items, and material deliveries into a one-page site summary. Every afternoon they type `/daily-report`, and the agent files the same report in the same format, with no need to re-explain what a good report looks like.

## Create a Skill

1. Open the agent's edit screen.
2. Switch to the **Skills** tab and click **New Skill**.
3. Give the Skill a slash-name, a short description, and the prompt or tool sequence.
4. Save. The Skill is available immediately to the agent and to anyone who uses it.

## How loading on demand works

Only the Skills relevant to the current task get loaded into the agent's working context. The rest stay parked, ready to fire when called by name. You can attach hundreds of Skills without bloating the agent's prompt or slowing it down.

Here is how a few common Skills behave once you attach them.

| Skill | When it loads | What it adds |
| --- | --- | --- |
| `/qa-checklist` | Task mentions "review" or "QA" | An 8-step quality pass on a draft |
| `/draft-followup` | A reply or follow-up is requested | A ready-to-send message in your tone |
| `/summarize-thread` | Asked to recap a long chat | A short bullet summary with action items |
| `/run-script` | A task needs a quick calculation or file step | A safe call into the [Bash Sandbox](https://www.taskade.com/learn/agents/bash-sandbox) |

## Persistence

Skills persist across reloads, agent restarts, and workspace shares. A teammate who uses your published agent gets the same set of slash-commands you do.

## Tips for great Skills

- **Name with intent:** use verbs like `/qa-checklist` or `/draft-followup`
- **Keep the prompt focused:** one clear goal per Skill is faster than a giant Swiss army knife
- **Pair with Tools:** call the [Bash Sandbox](https://www.taskade.com/learn/agents/bash-sandbox) or any built-in tool inside the Skill
- **Share the agent:** Skills travel with the agent when you publish or share it

## Related guides

- [Custom AI Agents](https://www.taskade.com/learn/agents/custom-agents). Build agents with specialized tools.
- [Bash Sandbox](https://www.taskade.com/learn/agents/bash-sandbox). Run scripts safely from inside chat.
- [Agent Tools](https://www.taskade.com/learn/agents/agent-tools). All the tools an agent can call.
- [Taskade EVE Mentions](https://www.taskade.com/learn/agents/eve-mentions). Reference workspace items in prompts.
- [Multi-Agents](https://www.taskade.com/learn/agents/multi-agents). Coordinate teams of agents.