Skills

A skill is a directory under skills/ containing a SKILL.md file. It bundles structured agent instructions, an allowed_tools policy, optional resource and reference files, static assets, and executable scripts. The format follows the agentskills.io specification.

Prerequisites

A Veryfront project with at least one agent (see Agents).
The skills/ directory exists at the project root, or ai.skills.discovery.paths is set in Configuration.

Quick start

Create a skill directory with a SKILL.md file:

skills/
  code-review/
    SKILL.md
    references/
      style-guide.md
    resources/
      review-rubric.md
    scripts/
      lint.sh

The SKILL.md file uses YAML frontmatter for metadata and Markdown for instructions:

---
name: code-review
description: Review code changes for style, correctness, and security issues.
allowed_tools: load_skill load_skill_reference execute_skill_script
---

# Code Review

Review the submitted code changes following the project style guide.

1. Load the style guide from `references/style-guide.md`
2. Load the rubric from `resources/review-rubric.md`
3. Check for common issues
4. Run the linter via `scripts/lint.sh`
5. Provide feedback with specific line references

Skill structure

Each skill lives in its own directory under skills/:

skills/<skill-id>/
├── SKILL.md              # Required: frontmatter + instructions
├── references/           # Optional: reference files the agent can read
│   └── *.md
├── resources/            # Optional: source documents or review inputs
│   └── *.md
├── scripts/              # Optional: executable scripts
│   └── *.sh
└── assets/               # Optional: static assets
    └── *

Frontmatter fields

Field	Required	Description
`name`	Yes	Skill identifier (lowercase alphanumeric + hyphens, 1-64 chars)
`description`	Yes	Human-readable description (max 1024 chars)
`allowed_tools`	No	Space-delimited tool IDs or prefix patterns (e.g. `api:*`) the agent may use
`license`	No	SPDX license identifier
`compatibility`	No	Compatibility constraints
`metadata`	No	Arbitrary key-value pairs

Discovery

Skills are discovered automatically from the skills/ directory at server startup and on HMR file changes. No registration is needed.

skills/
  code-review/SKILL.md     → skill ID: "code-review"
  data-analysis/SKILL.md   → skill ID: "data-analysis"

Agent tools

When skills are available, agents get three built-in tools:

Tool	Description
`load_skill`	Load a skill’s full instructions by ID
`load_skill_reference`	Read a file from `references/`, `resources/`, or `assets/`
`execute_skill_script`	Execute a script from a skill (5-minute timeout)

Enable skills on an agent:

// agents/assistant.ts
import { agent } from "veryfront/agent";

export default agent({
  id: "assistant",
  system: "Use project skills when they match the task.",
  skills: ["code-review"],
});

Expose the agent through an AG-UI route, then ask it to use the skill:

curl -N http://localhost:3000/api/ag-ui \
  -H "Content-Type: application/json" \
  -d '{"messages":[{"id":"1","role":"user","parts":[{"type":"text","text":"Use the code-review skill and summarize what you would check first."}]}]}'

The agent should call load_skill before applying the skill instructions.

Tool restrictions

The allowed_tools field restricts which tools an agent can use while a skill is active. Use exact IDs or prefix wildcards:

# Exact tool IDs
allowed_tools: load_skill load_skill_reference execute_skill_script

# Prefix wildcards
allowed_tools: api:* database:read

# No restriction (agent can use all tools)
# omit the field entirely

CLI commands

# Create a new skill
veryfront skills create my-skill

# Validate a skill
veryfront skills validate skills/my-skill

Verify it worked

Run veryfront skills validate skills/my-skill. A passing skill prints no errors and exits with status 0.
Restart veryfront dev. The dev log should list each registered skill under its directory name.
Send a message that should trigger the skill (for example, a code-review skill should engage when the message asks to “review this diff”). The AG-UI response should reference the skill’s instructions or call only the tools listed in allowed_tools.

Getting Started

Concepts

Guides

API Reference

Prerequisites

Quick start

Skill structure

Frontmatter fields

Discovery

Agent tools

Tool restrictions

CLI commands

Verify it worked

​Prerequisites

​Quick start

​Skill structure

​Frontmatter fields

​Discovery

​Agent tools

​Tool restrictions

​CLI commands

​Verify it worked

Prerequisites

Quick start

Skill structure

Frontmatter fields

Discovery

Agent tools

Tool restrictions

CLI commands

Verify it worked