---
slug: agent-skills-workflows-prompts-and-plays
title: Agent skills, workflows, prompts, and Plays compared
subtitle: Choose what to preserve for the next run
description: Compare prompts, agent skills, workflows, and Rote Plays by instructions, executable steps, dependencies, and sharing. Includes public examples and limitations.
author: Modiqo
published: 2026-08-28
updated: 2026-09-21
---

# Agent skills, workflows, prompts, and Plays compared

## Choose what to preserve for the next run

Choose a prompt for an immediate request, a skill for reusable guidance, and an executable procedure when the steps themselves need to travel. These approaches can work together. Their behavior depends on the implementation.

Rote by Modiqo turns successful AI agent work into reusable Plays. Inspect the steps, run them with new inputs, and share the method across supported agents.

## Compare what each approach preserves

| Approach | What it commonly contains | When it fits | What to check |
|---|---|---|---|
| Prompt | A request, context, and constraints | The task needs judgment or exploration | The model can interpret the request differently |
| Agent skill | Instructions, supporting files, and sometimes scripts | Guidance should apply across similar tasks | The skill can combine reasoning with executable code |
| Workflow | Steps, conditions, state, or a graph | A process needs explicit execution and maintenance | Workflows can be authored, generated, or derived from runs |
| Rote Play | A versioned procedure with inputs, requirements, and declared effects | A useful method should be inspected, reused, or shared | The runner still needs compatible tools, access, and valid inputs |

Skills can invoke scripts or Plays. Workflows can include model calls and adaptive branches. A Play can also use a model where its method requires one. None of these labels guarantees zero reasoning, lower cost, or identical output.

## A public Play makes the difference concrete

The [website readiness Play](https://play.modiqo.ai/modiqo/website-launch-readiness@1.1.3) checks a supplied URL. Its published inputs include `url`, `max_redirects`, and `run_lighthouse`. Its eight declared steps cover validation, DNS, fetching the site, TLS, robots, sitemap, optional Lighthouse, and the final assessment.

Inspect the method before checking a public site:

```bash
rote play inspect https://play.modiqo.ai/modiqo/website-launch-readiness@1.1.3
rote play run https://play.modiqo.ai/modiqo/website-launch-readiness@1.1.3 url=https://example.com run_lighthouse=false
```

The Play preserves the checks while the URL changes. A skill could help an agent choose this Play and explain the result. A larger workflow could use its result before a human approves a release.

This example needs network access and the tools listed in its source. Optional Lighthouse requires its local CLI and browser runtime. Its report can contain unavailable checks; a missing result is not evidence that a site passed.

## Choose based on the work you repeat

Keep an open-ended request as a prompt when the solution changes substantially each time. Use a skill when recurring work needs guidance, examples, and judgment. Use a workflow when your team needs to define or maintain the process directly.

Consider a Play when a useful method already exists and another person needs to inspect and run it. Review the version, inputs, effects, and dependencies. Compare actual results on your task before making a cost or reliability claim.

## Start with an existing procedure

[Trending Plays](/trending) includes published methods for DNS checks, service status, and repository review. Follow [the reuse example](/blog/how-an-agent-run-becomes-a-reusable-procedure), or learn [how to share a Play across agents](/blog/share-an-agent-workflow-without-sharing-credentials).