meain/fin/create-skill

Creating Agent Skills

Agent Skills are an open format (agentskills.io) for extending AI agents with specialized knowledge and workflows. A skill is a directory containing a SKILL.md file. Agents load skills progressively: only the name and description are read at startup; the full body loads only when the skill is activated.

Directory structure

skill-name/
├── SKILL.md          # Required: frontmatter + instructions
├── scripts/          # Optional: executable code the agent can run
├── references/       # Optional: documentation loaded on demand
├── assets/           # Optional: templates, data files, images
└── ...               # Any other files or directories

The directory name must match the name field in the frontmatter exactly.

SKILL.md frontmatter

---
name: skill-name
description: What the skill does and when to use it.
---

Two fields are required:

• `name` — 1–64 chars. Lowercase letters, digits, and hyphens only. No leading, trailing, or consecutive hyphens. Must match the directory name exactly.
• `description` — 1–1024 chars. Describes what the skill does and when to use it. This is the only thing the agent sees before deciding to activate.

Writing a good description

The description is the most important field. The agent reads it at startup to decide whether to activate the skill. Make it specific and trigger-oriented:

# Good — states what it does and when
description: Extracts text and tables from PDF files, fills PDF forms, and merges multiple PDFs. Use when working with PDF documents or when the user mentions PDFs, forms, or document extraction.
 
# Poor — too vague
description: Helps with PDFs.

Valid names: pdf-processing, data-analysis, code-review Invalid: PDF-Processing (uppercase), -pdf (leading hyphen), pdf--processing (consecutive hyphens)

Body

Everything after the closing --- is the skill body — plain markdown with no format restrictions. Write whatever helps the agent perform the task.

Recommended sections:

• Step-by-step instructions
• Input/output examples
• Common edge cases
• References to supplementary files (use relative paths from the skill root)

Keep SKILL.md under 500 lines (~5000 tokens). Move detailed reference material to separate files in references/ and instruct the agent to load them when needed.

Minimal example

.agents/skills/
  roll-dice/
    SKILL.md

---
name: roll-dice
description: Roll dice using a random number generator. Use when asked to roll a die (d6, d20, etc.) or generate a random dice roll.
---
 
To roll a die, run:
 
    echo $((RANDOM % <sides> + 1))
 
Replace `<sides>` with the number of sides (e.g. 6 for d6, 20 for d20).

Progressive disclosure

Skills load in three stages, from cheapest to most expensive:

1. Metadata (~100 tokens) — name + description, loaded for all skills at startup
2. Instructions (<5000 tokens recommended) — full SKILL.md body, loaded on activation
3. Resources (as needed) — files in scripts/, references/, assets/, loaded on demand

Structure your skill to take advantage of this: put just enough in the body to handle the common case, and reference supplementary files for edge cases or detailed lookup material.

Discovery locations

Agents typically discover skills from:

• .agents/skills/ walked up from the current working directory to the project root
• ~/.agents/skills/ for user-global skills

The first occurrence of a name wins, so project-level skills override user-level ones.