roebi New Work with Skills
Your Prompt in combination with an Agent is the new unit of work.
A practical guide to working with Agent Skills across increasing levels of mastery.
See references/levels.md for full detail on each level.
Quick Map of Levels
| Level | Title | Core Idea |
|---|---|---|
| 1 | Understand & Use Skills | A HowTo for Humans is a Skill for Agents |
| 2 | Save Single Work as a Skill | Capture sessions — don't let work get lost |
| 3 | Organize Your Skills | Git repos, flat folder, license, group freely |
| 4 | Continuous Skill Improvement | Live updates while you work |
| 5 | Skill with Special Techniques | Role skills, evolving state, TDD for skills |
| ∞ | This is the End of the Beginning | To be continued… |
Level 1 — Understand Skills and Use Skills
A HowTo for Humans is a Skill for Agents and Humans.
- What is a Skill: https://agentskills.io/specification
- A Skill can use other Skills → Skill Tree
- Care about Context Size: Read a Skill in 3 Steps
See: https://agentskills.io/what-are-skills#how-skills-work
The 3-Step Reading Pattern (Context-Efficient)
- Discovery — Agent loads only
name+description(~100 tokens per skill) - Activation — When task matches, agent reads full
SKILL.mdbody - Execution — Agent loads referenced files or executes scripts only as needed
Level 2 — Save Single Work as a Skill
The Problem: You worked on something. Your session ends. Your work is lost.
Old Solution: Save a summary as a .md file manually.
New Thinking: At the end of your session, use this prompt:
For this session write me a Skill named <my-new-skill> using
https://github.com/roebi/agent-skills/blob/main/skills/create-agent-skill-en/SKILL.md
Output is a Zip file ending with .zip that I can download.
It contains the new Skill and all its related files.
This turns every working session into a reusable, shareable, versioned Skill.
Level 3 — Organize Your Skills
- Store skills in one or more git repositories
- In a git repo, only a flat
skills/folder is allowed — no nested skill folders - Group by whatever fits: your team, your projects, your company
- License your skills — use the
licensefrontmatter field
→ Consider writing a dedicatedskill-licensorskill for consistent licensing
my-skills-repo/
└── skills/
├── my-skill-a/
│ └── SKILL.md
├── my-skill-b/
│ └── SKILL.md
└── new-work-with-skills-en/
└── SKILL.md
Level 4 — Continuous Skill Improvement
You work in your project git repo using an Agent, with your skills in a parallel folder <my-skills>.
While you work, when you notice a skill needs updating or extending, use this prompt pattern:
Update/Extend in the Skill <my-skill> this <whatever needs to change>.
This keeps skills alive and evolving alongside your actual work — without interrupting your flow.
Level 5 — Skill with Special Techniques
See references/special-techniques.md for full detail.
5a — Skill is a Role
Old Thinking: Agent has Roles. Agent uses Skills.
New Work Thinking:
Your Prompt + Agent → Let write a Skill that IS a Role,
using its own Skill, which may use other Skills.
The Role Skill can act as:
- Mentor — guides the process
- Safeguard — validates decisions
- Process Guard — enforces workflow
The Agent is reduced to pure EVA execution (Execute, Validate, Advance).
5b — Skill Having Evolving State
Instruct your Skill to output an evolving work result state .md file.
Loop:
- Skill reads your prompt + the evolving state file
- Skill processes
- Skill writes a new evolved state file
Repeat until the evolving work result state matches your expectations.
5c — Skill Until Requirement is Fulfilled
Think Test-Driven Development — implemented with Skills.
Write a Skill in generic form. Then combine it with:
- A simple Skill
- A Skill that is a Role (5a)
- A Skill having Evolving State (5b)
The skill loop runs until the defined requirement is fulfilled.
This is the End of the Beginning
To be continued…
Author: Robert Halter, Switzerland
Version: 2.0
Date: 2026-03-10
License: All rights reserved by author.