liveBook · Manning

Agent skills

As explained in Build a Multi-Agent System (from Scratch)

Agent Skills is an open standard for packaging reusable workflows that compliant LLM agent clients can discover, activate, and execute.

A recurring task, such as adding a newly discovered book to a reading list, can be captured once as a skill instead of being re-specified each time. Agent Skills is an open, framework-independent standard for packaging these reusable workflows so compatible LLM agent clients can discover and use them. section 6.0 section 1.5

Tools provide discrete functionality; a skill describes the procedure for completing a task, typically using one or more tools. A skill bundle contains a SKILL.md workflow and may include executable scripts, reference documents, and other assets. section 6.1.1 section 6.1.2

A skill bundle has these components:

  • SKILL.md: YAML frontmatter followed by the procedural workflow. The required frontmatter fields are name and description. section 6.1.2
  • Optional frontmatter: license, compatibility, metadata, and allowed-tools can provide additional information about use and requirements. section 6.1.2
  • Optional resources: scripts/, references/, and assets/ live alongside SKILL.md. The workflow can reference resources when they are needed; resource files should avoid deep nesting. section 6.1.2
Table 6.1 Fields that can be included in a skill’s frontmatter
Field
Required
Description
name
Yes
A unique identifier for the skill. Must be lowercase letters, numbers, and hyphens only. Cannot start or end with a hyphen. Maximum 64 characters.
description
Yes
Describes what the skill’s procedural workflow does and in what scenarios it should be applied. Maximum 1024 characters.
license
No
The name of the license that covers the skill (e.g. MIT) or the name of the license file that is bundled in the skills resources.
compatibility
No
Describes any environment constraints or requirements the skill has. Maximum 500 characters.
metadata
No
An arbitrary key-value map for additional properties not covered by the standard, like author or version.
allowed-tools
No
A space-delimited list of tools (each specified by name) the LLM agent can use to execute the skill.

The frontmatter field table gives the required fields, optional fields, and their constraints. section 6.1.2

Clients discover skill directories in project and user scope locations. For example, .agents/skills/ under the project directory or home directory is listed for all agent clients; clients may also use their own locations. Skills are progressively disclosed: after discovery, activation makes the workflow and bundled resources available, and the agent executes the procedure for the task. section 6.1.2 section 6.1.1

A discovery implementation can scan scope paths, validate each bundle, and return skills keyed by name. Validation can accept cosmetic issues with warnings while skipping fatal errors: an overlong name is cosmetic in this approach, whereas a missing SKILL.md or invalid frontmatter is fatal. section 6.3.2

Reading-list workflow

Document the repeated procedure for fetching information about a book from the web and adding it to a reading list. Once the skill is available to the agent, a user can request the task without restating the procedure. section 6.0

How Build a Multi-Agent System (from Scratch) covers it

How deep it goes
Specifies bundle structure and discovery, then implements validation and scanning with separate warnings and fatal errors.
Examples
Uses a web-assisted reading-list workflow to illustrate a repeatable skill.
What you can do after
Build support for discovering and validating skill bundles in an LLM agent framework.
How it connects
Connects skills with tool calling, Model Context Protocol, and progressive disclosure.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build a Multi-Agent System (from Scratch). The section links point to the parts of the book it draws on.