Agent Skills

Open standards

A skill is a directory of instructions, references, and optional scripts that an agent loads on demand. MAS-Lab skills follow the agentskills.io specification. The required SKILL.md has a short name and description, then detailed instructions that load only when the agent picks the skill.

How MAS-Lab implements it

mas-library-skills ships a ContextContract plugin and skill-access tools. The default engine is a bundled agentskills parser and sandbox (src/agentskills in that package). Optional extras wrap Google ADK and Deep Agents / LangChain; they are opt-in because those frameworks pin their own OpenTelemetry versions.

You attach a skill with an overlay. The agent manifest does not copy the instructions:

apiVersion: mas/v1
kind: Overlay
metadata:
  name: with-skills
spec:
  target:
    kind: Agent
  patch:
    spec:
      skills:
        - name: answer-expert
          path: skills/answer-expert

Dependencies

Piece	Package
Plugin, parser, sandbox	mas-library-skills
Specification	agentskills.io
Optional engines	extras adk (google-adk) and langchain (deepagents)

Compliance

Skills are valid when SKILL.md meets the agentskills.io front matter and layout. MAS-Lab does not run a third-party skills TCK. Tutorial 6's answer-expert tree is the in-repo fixture:

docs/tutorials/06-agent-skills/skills/answer-expert/
├── SKILL.md
├── references/physical-constants.md
└── scripts/convert_speed.py

Related

• Tutorial 6 — Agent skills
• Agent manifest (spec.skills)
• Package map