Artificial Intelligence (AI) Agents Skills

About AI Agent Skills

Source: What AI Agent Skills Are and How They Work - IBM - YouTube with Martin Keen

Use case: AI agents are good at getting information, but weak at procedural knowledge like creating a financial report using 30 steps or how to ride a motorcycle in Rome Italy safely. Skills address this weakness by giving the agent the process to follow.

Skills answer what to do, the order to do them, and judgment. They teach an agent how to do a job (a repeatable task).

Creating a Skill

Creating a skill is creating a SKILL.md markdown file with name, description describing when the skill applies.

---
name: pdf-builder
description: Use when the user asks to extract a PDF
---

# PDF Builder

Instructions to agent, for example read content, run command to build PDF

Optional skill folders:

/scripts/
- Programming language files agent can run like Python, bash, Java
/references/
- Documentation
/assets/
- Data files
- Templates

Loading Skills using Progressive Disclosure

Agent is given name and description of skills it has access to. Like an index of skills. This approach reduces context needed (reduce token use).

If agent finds a description that is useful, it will load the skill.md. Optional folders are only loaded when a task is required.

Knowledge: MCP, RAG, Fine tuning, and Skills

Model Context Protocol (MCP) - gives tool access
- Context engineering with give an agent MCP (tools and data access) and skills (steps to do specific tasks). See context engineering at Artificial Intelligence (AI) Agents and Model Context Protocol (MCP) - Artificial Intelligence (AI) Agents and Model Context Protocol (MCP)
Retrieval Augmented Generation (RAG) - facts, knowledge
Fine tuning - adjusts model weights
Skills - procedural knowledge, when to use other knowledge

Skills.md format

Open standard published at Agent Skills Overview - Agent Skills

my-skill/
├── SKILL.md          # Required: metadata + instructions
├── scripts/          # Optional: executable code
├── references/       # Optional: documentation
├── assets/           # Optional: templates, resources
└── ...               # Any additional files or directories

Memory

Semantic memory –> RAG

Episodic memory –> Conversation logs, interaction history

Procedural memory –> Skill files

Trust and Skills

Skills can have access to scripts that run on the user’s machine, environment variables, and secrets. Review skills before using them.